# function calling

A technique that lets an AI model identify a function to invoke and provide the arguments needed to execute it, connecting natural-language input with programmatic functions.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## Five Guardrails That Limit the Damage an AI Agent Can Do

DevFeed: [Five Guardrails That Limit the Damage an AI Agent Can Do](<https://devfeed.tech/articles/five-guardrails-that-limit-the-damage-an-ai-agent-can-do-145977.md>)

Original publisher: [Read original article](<https://towardsdatascience.com/five-guardrails-that-limit-the-damage-an-ai-agent-can-do/>)

Author: Thomas Reid

Published: 2026-11-20T12:00:00Z

Content type: tutorial

Language: en

Sources: [Towards Data Science](<https://devfeed.tech/sources/towards-data-science.md>)

Topics: [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [Security](<https://devfeed.tech/topics/security.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>), [Test-driven development](<https://devfeed.tech/topics/tdd.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [arguments](<https://devfeed.tech/tags/arguments.md>), [guardrail](<https://devfeed.tech/tags/guardrail.md>), [guardrails](<https://devfeed.tech/tags/guardrails.md>), [llmops](<https://devfeed.tech/tags/llmops.md>), [pydantic](<https://devfeed.tech/tags/pydantic.md>), [pytest](<https://devfeed.tech/tags/pytest.md>), [python](<https://devfeed.tech/tags/python.md>), [sqlite](<https://devfeed.tech/tags/sqlite.md>), [tests](<https://devfeed.tech/tags/tests.md>)

### AI overview

The article's central safeguard is to let Python enforce what an AI agent can do, since prompts alone cannot reliably prevent harmful actions. It treats user questions, model tool calls, and returned data as untrusted, including indirect prompt injection in database notes. The example uses Pydantic models to reject unknown fields and constrain arguments, an allowlisted dispatcher for read tools, read-only SQLite connections with limited query results, and a cap on the agent's turns. A return write stays outside the autonomous tool loop and is saved only after user confirmation; a UUID idempotency key and database uniqueness constraint prevent duplicate requests on retry. Pytest checks the deterministic controls. The example agent ignored Mina's malicious note, but the article cautions that this does not fully mitigate prompt injection; application-enforced limits reduce the damage a bad model decision can cause.

### Source excerpt

AI Agents for Beginners, Part three: strict Pydantic arguments, a tool allowlist, read-only SQLite, a step cap, and confirmed writes, all verified by five pytest tests in 0.68s. The post Five Guardrails That Limit the Damage an AI Agent Can Do appeared first on Towards Data Science.

## Living off the Local Model: Abusing Browser-Embedded Language Models for Post-Exploitation

DevFeed: [Living off the Local Model: Abusing Browser-Embedded Language Models for Post-Exploitation](<https://devfeed.tech/articles/living-off-the-local-model-abusing-browser-embedded-language-models-for-post-exploitation-145464.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25086>)

Author: William Fleshman, Joseph Lucas

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [LLM security](<https://devfeed.tech/topics/llm-security.md>), [Local AI](<https://devfeed.tech/topics/local-ai.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Google Chrome](<https://devfeed.tech/topics/google-chrome.md>), [Cryptography](<https://devfeed.tech/topics/cryptography.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>)

Tags: [browser](<https://devfeed.tech/tags/browser.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [embedded](<https://devfeed.tech/tags/embedded.md>), [exploitation](<https://devfeed.tech/tags/exploitation.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [local](<https://devfeed.tech/tags/local.md>), [post-exploitation](<https://devfeed.tech/tags/post-exploitation.md>)

### AI overview

The paper reports that Chrome's on-device Gemini Nano can be driven from the host using operating-system interpreters, and that its one-time user-activation check for downloading the model can be met with a simulated click or bypassed. The authors test credential identification, PII detection in email, image classification, and custom tool use. In a simulated operation, they combine these capabilities to read a password from a webcam image, unlock and triage an encrypted archive, and stage a private key for exfiltration without user interaction. These findings describe the current model and a simulated scenario.

### Source excerpt

arXiv:2611.25086v1 Announce Type: cross Abstract: Browsers now ship local language models to ordinary endpoints. Google includes Gemini Nano in Chrome to enable productivity features such as summarization, with inference running entirely on the device. These browser-embedded models provide new living-off-the-land opportunities for attackers, each instance a potential post-exploitation engine already installed, signed by the vendor, and invisible to network monitoring. The Gemini model is multimodal, capable of processing files, analyzing screenshots, or conversing with an attacker. We show that it can be driven from the host using only interpreters that ship with the operating system and find that the one-time user-activation check before the model is downloaded is easily satisfied with a simulated click or bypassed entirely.We measure and report the current model's ability to identify credentials on the host, detect PII in email, classify images, and call custom tools. We combine these capabilities and use the model to carry out a simulated cyber operation: reading a password from a webcam image, unlocking an encrypted archive, triaging the contents, and staging a private key for exfiltration, end-to-end without user-interaction.

## LLM4Rail: An LLM-Augmented Personalized Railway Services Platform

DevFeed: [LLM4Rail: An LLM-Augmented Personalized Railway Services Platform](<https://devfeed.tech/articles/llm4rail-an-llm-augmented-personalized-railway-services-platform-145306.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2507.23377>)

Author: Zhuo Li, Xianghuai Deng, Chiwei Feng, Hanmeng Li, Shenjie Wang, Junchao Fan, Jiaxing Shen, Zhiyuan Wen, Longxiang Gao, Linchun Wu

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [recommendation systems](<https://devfeed.tech/topics/recommendation-systems.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>)

Tags: [alignment](<https://devfeed.tech/tags/alignment.md>), [arguments](<https://devfeed.tech/tags/arguments.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [food](<https://devfeed.tech/tags/food.md>), [framework](<https://devfeed.tech/tags/framework.md>), [generation](<https://devfeed.tech/tags/generation.md>), [generative](<https://devfeed.tech/tags/generative.md>), [language](<https://devfeed.tech/tags/language.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llms](<https://devfeed.tech/tags/llms.md>), [mechanism](<https://devfeed.tech/tags/mechanism.md>), [modules](<https://devfeed.tech/tags/modules.md>), [novel](<https://devfeed.tech/tags/novel.md>), [platform](<https://devfeed.tech/tags/platform.md>), [railway](<https://devfeed.tech/tags/railway.md>), [service](<https://devfeed.tech/tags/service.md>), [services](<https://devfeed.tech/tags/services.md>), [similarity](<https://devfeed.tech/tags/similarity.md>), [system](<https://devfeed.tech/tags/system.md>)

### AI overview

LLM4Rail is a personalized railway-services platform with modules for ticket and weather inquiries, onboard food and drink recommendations, and casual conversation. Its iterative ER-QTAO framework uses Experience Storage and Retrieval and Diagnostic Observation Generation to help the LLM call external tools with complex arguments. For catering, the authors introduce the Chinese Railway Food & Drink dataset, organized by city, cuisine, age group, and spiciness, and a zero-shot conversational recommender. A feature-similarity alignment mechanism is intended to keep generated recommendations matched to dataset entries.

### Source excerpt

arXiv:2507.23377v2 Announce Type: replace Abstract: Large language models (LLMs) have significantly reshaped different walks of business. To meet the increasing demands for individualized railway service, we develop LLM4Rail - a novel LLM-augmented personalized railway services platform. Empowered by the LLM, LLM4Rail provides dedicated modules for ticket and weather inquiries, food & drink recommendations, and casual conversation. In LLM4Rail, we propose the iterative "Experience-Retrieval Question-Thought-Action-Observation (ER-QTAO)" reasoning framework. It introduces two key mechanisms, Experience Storage and Retrieval (ESR) and Diagnostic Observation Generation (DOG), to enhance the LLM's ability to invoke appropriate external tools with complex arguments. To provide personalized onboard dining services, we first construct the Chinese Railway Food & Drink (CRFD) dataset, which covers a wide range of signature dishes categorized by city, cuisine, age group, and spiciness level, and then develop an LLM-based zero-shot conversational recommender system for railway catering. In addition, to address the unconstrained nature of generative recommendations, we design a feature-similarity-based alignment mechanism to ensure that all recommended items are aligned with entries in the CRFD dataset.

## Duopra: From Full-Duplex Conversation to Full-Duplex Agency

DevFeed: [Duopra: From Full-Duplex Conversation to Full-Duplex Agency](<https://devfeed.tech/articles/duopra-from-full-duplex-conversation-to-full-duplex-agency-145586.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25773>)

Author: Chiyu Chen, Pengzhou Cheng, Yuezhao Ma, Xiaocheng Li, Can Liu, Haonan Hu, Yao Yao, Yifei Yang, Yijie Lu, Taolei Wang, Yansi Li, Muhuan Wu, Chenglu Sun, Gongshen Liu, Zhuosheng Zhang

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>), [post-training](<https://devfeed.tech/topics/post-training.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>)

Tags: [audio](<https://devfeed.tech/tags/audio.md>), [backend](<https://devfeed.tech/tags/backend.md>), [bench](<https://devfeed.tech/tags/bench.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [execution](<https://devfeed.tech/tags/execution.md>), [frontend](<https://devfeed.tech/tags/frontend.md>), [full-duplex-bench](<https://devfeed.tech/tags/full-duplex-bench.md>), [interaction](<https://devfeed.tech/tags/interaction.md>), [interface](<https://devfeed.tech/tags/interface.md>), [latency](<https://devfeed.tech/tags/latency.md>), [low-latency](<https://devfeed.tech/tags/low-latency.md>), [model](<https://devfeed.tech/tags/model.md>), [natural-language](<https://devfeed.tech/tags/natural-language.md>), [pass](<https://devfeed.tech/tags/pass.md>), [pipeline](<https://devfeed.tech/tags/pipeline.md>), [protocol](<https://devfeed.tech/tags/protocol.md>), [realtime](<https://devfeed.tech/tags/realtime.md>), [repair](<https://devfeed.tech/tags/repair.md>), [request](<https://devfeed.tech/tags/request.md>), [requires](<https://devfeed.tech/tags/requires.md>), [v3](<https://devfeed.tech/tags/v3.md>)

### AI overview

Duopra addresses a gap in full-duplex agents: fluent turn-taking does not ensure timely, complete tool requests or corrections. It treats the backend as a tool that receives a self-contained natural-language request, using one action interface for direct calls and backend delegation. Training combines protocol SFT, full-duplex reinforcement learning for timely and complete execution, and projector-only repair of talker conditioning after thinker updates. On FDB v3, the authors report Pass@1 rising from 4.7% to 44.7% with the same backend; Duopra reaches 47.7% with direct tool calls (26.0% after SFT) and 49.3% with a DeepSeek backend. It also improves three FDB v1.0 takeover rates over the base model. The excerpt supports the authors' claim that task agency can coexist with key conversational agency.

### Source excerpt

arXiv:2611.25773v1 Announce Type: new Abstract: Full-duplex interaction models exhibit strong conversational agency, enabling natural, low-latency turn-taking, backchannels, and interruptions. However, completing tasks further requires task agency: timely tool invocation and argument revision following user corrections. Frontend-backend agents separate initiation from execution: the frontend decides whether and what to request, while the backend determines how. Execution can be offloaded, but unsent requests cannot be executed, and incomplete ones can misdirect the backend. On Full-Duplex-Bench v3 (FDB v3), most failures of Realtime-Venus-Audio + Harness in our streaming pipeline occur before a domain tool executes. We introduce Duopra, a full-duplex interaction model that learns task execution alongside conversational control. Duopra treats the backend as a tool, accepting a self-contained natural-language request. Direct calls and backend delegation share an action interface and requirements for faithful initiation and correct execution. Its three-stage training combines protocol SFT, full-duplex RL rewarding timely and complete execution, and projector-only repair of talker conditioning after thinker updates. With the same backend, Pass@1 rises from 4.7% to 44.7%; Duopra achieves 47.7% calling tools directly (26.0% after SFT) and 49.3% with a DeepSeek backend. It improves three FDB v1.0 takeover rates over the base model. These results show that task agency can coexist with key conversational agency.

## What Is Tool Calling? How LLMs and AI Agents Take Action

DevFeed: [What Is Tool Calling? How LLMs and AI Agents Take Action](<https://devfeed.tech/articles/what-is-tool-calling-how-llms-and-ai-agents-take-action-145886.md>)

Original publisher: [Read original article](<https://montecarlo.ai/blog-what-is-tool-calling>)

Author: Virna Sekuj

Published: 2026-11-19T16:58:24Z

Content type: article

Language: en

Sources: [Monte Carlo](<https://devfeed.tech/sources/monte-carlo.md>)

Topics: [function calling](<https://devfeed.tech/topics/function-calling.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [LLM observability](<https://devfeed.tech/topics/llm-observability.md>), [API](<https://devfeed.tech/topics/api.md>), [Tool](<https://devfeed.tech/topics/tool.md>)

Tags: [2023](<https://devfeed.tech/tags/2023.md>), [2026](<https://devfeed.tech/tags/2026.md>), [actions](<https://devfeed.tech/tags/actions.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-behavior](<https://devfeed.tech/tags/agent-behavior.md>), [agent-frameworks](<https://devfeed.tech/tags/agent-frameworks.md>), [agent-trust](<https://devfeed.tech/tags/agent-trust.md>), [agent-trust-glossary](<https://devfeed.tech/tags/agent-trust-glossary.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [api](<https://devfeed.tech/tags/api.md>), [database](<https://devfeed.tech/tags/database.md>), [function](<https://devfeed.tech/tags/function.md>), [function-calling](<https://devfeed.tech/tags/function-calling.md>), [glossary](<https://devfeed.tech/tags/glossary.md>), [large-language-model](<https://devfeed.tech/tags/large-language-model.md>), [llm-tool-calling](<https://devfeed.tech/tags/llm-tool-calling.md>), [llms](<https://devfeed.tech/tags/llms.md>), [querying](<https://devfeed.tech/tags/querying.md>), [searching](<https://devfeed.tech/tags/searching.md>), [structured](<https://devfeed.tech/tags/structured.md>), [tool-calling](<https://devfeed.tech/tags/tool-calling.md>), [tool-calling-examples](<https://devfeed.tech/tags/tool-calling-examples.md>), [tool-calling-vs-mcp](<https://devfeed.tech/tags/tool-calling-vs-mcp.md>), [tool-use-ai-agents](<https://devfeed.tech/tags/tool-use-ai-agents.md>), [uses](<https://devfeed.tech/tags/uses.md>)

### AI overview

Tool calling lets an LLM request an action by returning a structured tool name and arguments; the application validates and executes the request, sends back the result, and the model uses it to answer or make another call. Function calling and tool use are alternate names for the same mechanism. MCP standardizes how compatible applications discover and connect to tools; it builds on tool calling rather than replacing it. Calls can fail through wrong tool choice, bad arguments, skipped checks, retries, mishandled errors, changed responses, or unintended access, often without obvious errors. The text recommends tracing each call and its result, validating arguments and permissions before execution, watching for unexpected patterns and loops, tracking cost and latency, monitoring the underlying data, and limiting or requiring approval for high-impact actions.

### Source excerpt

Tool calling, also called function calling, is how a large language model asks an application to run a function, such as querying a database, calling an API, or searching documents, and then uses the result in its answer. The model doesn't run the tool itself. It produces a structured request naming the tool and its arguments, the application executes it, and the result goes back to the model.

## Build Tool-Calling Parsers in Rust with AI Agents

DevFeed: [Build Tool-Calling Parsers in Rust with AI Agents](<https://devfeed.tech/articles/build-tool-calling-parsers-in-rust-with-ai-agents-143887.md>)

Original publisher: [Read original article](<https://developer.nvidia.com/blog/build-tool-calling-parsers-in-rust-with-ai-agents/>)

Author: Keiven Chang

Published: 2026-11-19T16:30:46Z

Content type: tutorial

Language: en

Sources: [NVIDIA Developer](<https://devfeed.tech/sources/nvidia-developer.md>), [NVIDIA Technical Blog](<https://devfeed.tech/sources/nvidia-technical-blog.md>)

Topics: [function calling](<https://devfeed.tech/topics/function-calling.md>), [rust-ai](<https://devfeed.tech/topics/rust-ai.md>), [Test-driven development](<https://devfeed.tech/topics/tdd.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [Nemotron](<https://devfeed.tech/topics/nemotron.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [3](<https://devfeed.tech/tags/3.md>), [adapter](<https://devfeed.tech/tags/adapter.md>), [adapting](<https://devfeed.tech/tags/adapting.md>), [agentic-ai-generative-ai](<https://devfeed.tech/tags/agentic-ai-generative-ai.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [argument](<https://devfeed.tech/tags/argument.md>), [build](<https://devfeed.tech/tags/build.md>), [call](<https://devfeed.tech/tags/call.md>), [can](<https://devfeed.tech/tags/can.md>), [developer-tools-techniques](<https://devfeed.tech/tags/developer-tools-techniques.md>), [dynamo](<https://devfeed.tech/tags/dynamo.md>), [frontend](<https://devfeed.tech/tags/frontend.md>), [inference-performance](<https://devfeed.tech/tags/inference-performance.md>), [leak](<https://devfeed.tech/tags/leak.md>), [markup](<https://devfeed.tech/tags/markup.md>), [model](<https://devfeed.tech/tags/model.md>), [rust](<https://devfeed.tech/tags/rust.md>), [tool](<https://devfeed.tech/tags/tool.md>), [tool-calling](<https://devfeed.tech/tags/tool-calling.md>), [turns](<https://devfeed.tech/tags/turns.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

Tool-call parsers must separate text, reasoning, and calls while preserving streaming boundaries and avoiding lost data, duplicate calls, or leaked model markup. The article's Rust workflow uses frontend-crates and its shared Conformance Corpus: adapt applicable UNIFIED fixtures, create and human-review GOLDEN expectations against the model specification and related sources, then implement the parser and inspect captured results in the local conformance report. An unavailable tool is reported as generated; the serving layer checks availability, authorization, schema, and policy. Incomplete calls require explicit buffering and end-of-stream recovery rules rather than guessing missing content. CPU fixtures suffice for parser conformance, while end-to-end validation needs Dynamo runtime, discovery services, and model hardware. Further checks should measure speed and memory across input and chunk sizes, verify incremental event delivery and truncation behavior, and test the configured quote-handling policy. Guided decoding may reduce malformed calls but cannot prevent truncation or incorrect tool choices.

### Source excerpt

A tool call can arrive in pieces, an argument can be cut short, or model-specific markup can leak into the answer. A frontend parser turns that output, received...

## Robustness of Text-Only Adapted Speech LLMs for Spoken Semantic Parsing

DevFeed: [Robustness of Text-Only Adapted Speech LLMs for Spoken Semantic Parsing](<https://devfeed.tech/articles/robustness-of-text-only-adapted-speech-llms-for-spoken-semantic-parsing-142533.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2606.21453>)

Author: Youngwon Choi, Donghyuk Jung, Hyeonyu Kim, Taeyoun Kwon, Myeongkyun Cho, Moon Hwan Kim

Published: 2026-11-19T05:00:00Z

Content type: news

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Large Language Model](<https://devfeed.tech/topics/llm.md>), [asr](<https://devfeed.tech/topics/asr.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>)

Tags: [asr](<https://devfeed.tech/tags/asr.md>), [command](<https://devfeed.tech/tags/command.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-hc](<https://devfeed.tech/tags/cs-hc.md>), [cs-sd](<https://devfeed.tech/tags/cs-sd.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [eess-as](<https://devfeed.tech/tags/eess-as.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [function-calling](<https://devfeed.tech/tags/function-calling.md>), [inference](<https://devfeed.tech/tags/inference.md>), [llms](<https://devfeed.tech/tags/llms.md>), [noise](<https://devfeed.tech/tags/noise.md>), [parsing](<https://devfeed.tech/tags/parsing.md>), [performance](<https://devfeed.tech/tags/performance.md>), [robustness](<https://devfeed.tech/tags/robustness.md>), [semantic](<https://devfeed.tech/tags/semantic.md>), [speech](<https://devfeed.tech/tags/speech.md>), [text](<https://devfeed.tech/tags/text.md>)

### AI overview

Across matched text-only supervision, Speech LLMs generally outperform ASR-LLM cascades on spoken command understanding, intent prediction, and function calling, especially under severe babble noise. The study compares 3B and 7B models on three datasets using clean speech and five noise SNR levels. Gains on fine-grained entity prediction are less consistent. Speech LLMs also have lower average inference time per sample at both model scales. These findings are limited to the evaluated tasks, datasets, and conditions.

### Source excerpt

arXiv:2606.21453v2 Announce Type: replace-cross Abstract: Large language models (LLMs) are increasingly used to generate structured outputs from natural language instructions, including semantic frames and function calls. When the input is speech, this task becomes spoken semantic parsing, which can be performed using either Speech LLMs or ASR-LLM cascades. Although both approaches can be adapted with text-only task supervision, their relative robustness to acoustic degradation remains unclear. We compare these approaches under matched text-form supervision using 3B and 7B models across three datasets, evaluated on clean speech and babble noise at five SNR levels. Results show that Speech LLMs generally achieve better performance in command understanding, intent prediction, and function calling, particularly under severe noise, while improvements in fine-grained entity prediction are less consistent. An additional evaluation shows that Speech LLMs achieve lower average inference time per sample than ASR-LLM cascades at both model scales.

## Beyond the Parameter Monolith: Reconstructive Memories, Executable Skills, and Residual Assembly for Language Models

DevFeed: [Beyond the Parameter Monolith: Reconstructive Memories, Executable Skills, and Residual Assembly for Language Models](<https://devfeed.tech/articles/beyond-the-parameter-monolith-reconstructive-memories-executable-skills-and-residual-assembly-for-language-models-137289.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.04012>)

Author: A. Bochkov

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [AI Integration Strategies](<https://devfeed.tech/topics/ai-integration-strategies.md>), [Retrieval Augmented Generation (RAG)](<https://devfeed.tech/topics/retrieval-augmented-generation-rag.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>)

Tags: [adapters](<https://devfeed.tech/tags/adapters.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [artifacts](<https://devfeed.tech/tags/artifacts.md>), [assembly](<https://devfeed.tech/tags/assembly.md>), [bounded](<https://devfeed.tech/tags/bounded.md>), [codec](<https://devfeed.tech/tags/codec.md>), [consistency](<https://devfeed.tech/tags/consistency.md>), [construction](<https://devfeed.tech/tags/construction.md>), [contracts](<https://devfeed.tech/tags/contracts.md>), [core](<https://devfeed.tech/tags/core.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [execution](<https://devfeed.tech/tags/execution.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [external](<https://devfeed.tech/tags/external.md>), [field](<https://devfeed.tech/tags/field.md>), [fixed](<https://devfeed.tech/tags/fixed.md>), [independent](<https://devfeed.tech/tags/independent.md>), [language](<https://devfeed.tech/tags/language.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [skills](<https://devfeed.tech/tags/skills.md>)

### AI overview

FEM-ASM lets independently produced document memories and bounded executable-skill results participate in shared neural computation without jointly optimizing every component. Workers create immutable, versioned memories using a frozen codec; type-specific adapters, explicit support mappings, and a shared residual-assembly operator connect memories and execution results to the neural field. The computation combines a quadratic consistency step with a recurrent learned correction. Experiments with unchanged core weights distinguish recoverable storage, successful retrieval, and successful neural use as separate properties. Position-specific observations improved the tested memory and execution-result interfaces, and attaching new evidence through the existing interface affected predictions. The supplied abstract describes the architecture and experiments but does not provide quantitative results.

### Source excerpt

arXiv:2610.04012v2 Announce Type: replace Abstract: Can a language-model system grow its external state and use executable procedures without making every component part of one shared optimization process? We present FEM-ASM, an implemented architecture that separates neural coordination, document-state construction, and deterministic execution. Independent workers optimize document memories under a frozen codec and publish immutable, versioned artifacts. Executable skills return bounded computation results. Both enter a shared neural field through type-specific adapters, explicit support mappings, and a common residual-assembly operator. The resulting computation couples a quadratic consistency substep to a recurrent learned correction. We evaluate memory construction, controlled retrieval, positional readout, execution-result rendering, and evidence attachment with unchanged core weights. The experiments show that recoverable storage, successful retrieval, and successful neural use are distinct properties. Position-specific observations improve the tested memory and execution-result interfaces, while fixed-core attachment demonstrates that newly available evidence can affect predictions through the existing interface. The contribution is an architectural and engineering account of independently produced state participating in shared neural computation, with explicit component contracts, executable protocols, and released research artifacts.

## Auditable Authorization for LLM Agents: Empirical Evidence Across Model, Harness, and Workflow Layers

DevFeed: [Auditable Authorization for LLM Agents: Empirical Evidence Across Model, Harness, and Workflow Layers](<https://devfeed.tech/articles/auditable-authorization-for-llm-agents-empirical-evidence-across-model-harness-and-workflow-layers-137422.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19237>)

Author: Bangjie Deng, Xiaoxuan Lu, Wenjun Chen, Chengge Wen, Yulin Wang, Jiayue Meng, Yao Yu, Liying Liang, Weijia Han, Ziwei Hong, Zidong Yu

Published: 2026-11-17T05:00:00Z

Content type: news

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Authorization](<https://devfeed.tech/topics/authorization.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [audit](<https://devfeed.tech/topics/audit.md>), [agentic workflows](<https://devfeed.tech/topics/agentic-workflows.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>), [n8n](<https://devfeed.tech/topics/n8n.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [audit](<https://devfeed.tech/tags/audit.md>), [authorization](<https://devfeed.tech/tags/authorization.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [harness](<https://devfeed.tech/tags/harness.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-agents](<https://devfeed.tech/tags/llm-agents.md>), [model](<https://devfeed.tech/tags/model.md>), [workflow](<https://devfeed.tech/tags/workflow.md>)

### AI overview

The audit finds that approval controls can fail in two ways: rejected work may still execute, or approved work may not complete. It examines three boundaries--the model's call selection, the harness's scheduling and resumption, and the workflow's routing to protected effects--using paired decisions, execution traces, and business-state evidence. Reported results include seven unfinished approvals out of 300 with no denied effects in the paired model panel; lost approved branches under parallel confirmation; 336 of 672 CrewAI approvals omitted during restoration; and execution after rejection in four of 59 comparable n8n workflows. Comparing intent, generated calls, dispatch, and resulting effects is presented as a way to locate the first observable divergence and its repair target. The authors propose an audit contract to verify refusal safety and bounded authorization completeness separately.

### Source excerpt

arXiv:2611.19237v1 Announce Type: new Abstract: Human approval provides effective control only if rejected work stops and authorized work completes. We formalize refusal safety and bounded authorization completeness, and audit three execution boundaries: the model selects a call, the harness schedules and resumes it, and the workflow routes the decision to a protected effect. Paired decisions, exact traces, and native business-state witnesses support experiments with three models, five prompt arms, four runtime approval designs, two additional framework replays, and 100 public workflow sources. The paired model panel leaves seven of 300 approvals unfinished despite zero denied effects; parallel confirmation loses an approved branch in three tasks per model, restoration omits 336/672 CrewAI approvals, and four of 59 comparable n8n workflows execute after Reject. Comparing intent, generated calls, dispatch, and native effects identifies the first observable divergence and its repair target. This yields a practical audit contract that makes both refusal safety and authorization completeness independently verifiable, converting human approval from a procedural formality into an enforceable control boundary between intent and real-world effect.

## AMTFV: Agentic Mathematical Tool-Flow Verification for LLM Self-Correction

DevFeed: [AMTFV: Agentic Mathematical Tool-Flow Verification for LLM Self-Correction](<https://devfeed.tech/articles/amtfv-agentic-mathematical-tool-flow-verification-for-llm-self-correction-137269.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2607.29549>)

Author: Rui Zou, Yutao Zhu, Mengqi Wei, Ji-Rong Wen

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [math-verify](<https://devfeed.tech/topics/math-verify.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [deepseek](<https://devfeed.tech/tags/deepseek.md>), [execution](<https://devfeed.tech/tags/execution.md>), [flow](<https://devfeed.tech/tags/flow.md>), [gemini](<https://devfeed.tech/tags/gemini.md>), [gpt](<https://devfeed.tech/tags/gpt.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [llm](<https://devfeed.tech/tags/llm.md>), [mathematical-reasoning](<https://devfeed.tech/tags/mathematical-reasoning.md>), [model](<https://devfeed.tech/tags/model.md>), [modeling](<https://devfeed.tech/tags/modeling.md>), [natural-language](<https://devfeed.tech/tags/natural-language.md>), [problem-solving](<https://devfeed.tech/tags/problem-solving.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [reflection](<https://devfeed.tech/tags/reflection.md>), [request](<https://devfeed.tech/tags/request.md>), [tool](<https://devfeed.tech/tags/tool.md>), [verification](<https://devfeed.tech/tags/verification.md>)

### AI overview

AMTFV uses a Mathematical Tool Flow (MTF) interrupt-execute-resume interface to separate verification planning from exact computation. A verification agent builds a workflow and sends mathematical objects and computational intent to a toolbox agent, which generates executable calls for a backend; the results inform answer adjudication and revisions to both the answer and workflow. The authors report evaluation on five mathematical reasoning datasets and seven DeepSeek, GPT, and Gemini configurations. AMTFV outperformed the study's representative baselines overall, with average accuracy gains of up to 8.3 percentage points over the strongest baseline in an individual configuration, and larger gains on medium- and high-verification-complexity samples.

### Source excerpt

arXiv:2607.29549v3 Announce Type: replace Abstract: Large language models have demonstrated strong mathematical problem-solving capabilities, yet reliably verifying their candidate answers remains challenging. Existing representative methods mainly revise outputs through natural-language reflection or assist verification by directly generating verification programs; the former may not reliably support exact computation, whereas the latter prematurely couples mathematical modeling with low-level implementation. We propose AMTFV (Agentic Mathematical Tool-Flow Verification). By introducing Mathematical Tool Flow (MTF) as an interrupt--execute--resume interface, AMTFV decouples verification modeling from concrete execution and supports exact computation through a mathematical toolbox. Specifically, the verification agent first constructs a verification workflow, encodes the mathematical objects and computational intent requiring reliable execution in an MTF request, and sends it to the mathematical toolbox agent. The latter parses the request, generates executable calls, and dispatches them to the backend for exact computation. Tool outputs then support candidate-answer adjudication, answer revision, and verification-workflow revision. We evaluate AMTFV on five challenging mathematical reasoning datasets with seven model configurations from DeepSeek, GPT, and Gemini. Experimental results show that AMTFV outperforms the representative baselines evaluated in this study overall; under an individual model configuration, it improves average accuracy over the strongest baseline by up to 8.3 percentage points, with larger gains on samples of medium and high verification complexity.

## RGFC: Registry Grounded Function Calling with stage-decomposed evaluation

DevFeed: [RGFC: Registry Grounded Function Calling with stage-decomposed evaluation](<https://devfeed.tech/articles/rgfc-registry-grounded-function-calling-with-stage-decomposed-evaluation-137968.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.18622>)

Author: Ritvik Garimella, Vedant Khandelwal, Anvi Kohli, Amit Sheth

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Software Engineering](<https://devfeed.tech/sources/arxiv-software-engineering.md>)

Topics: [function calling](<https://devfeed.tech/topics/function-calling.md>)

Tags: [benchmark](<https://devfeed.tech/tags/benchmark.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-se](<https://devfeed.tech/tags/cs-se.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [function-calling](<https://devfeed.tech/tags/function-calling.md>), [grounding](<https://devfeed.tech/tags/grounding.md>), [registry](<https://devfeed.tech/tags/registry.md>), [schema](<https://devfeed.tech/tags/schema.md>), [tool-calling](<https://devfeed.tech/tags/tool-calling.md>)

### AI overview

The authors introduce RFCD, a benchmark with standardized JSON-schema tool registries ranging from 5 to 3,000 functions, and TAAG, a deterministic evaluator that separates registry conformance, structural completeness, and argument grounding. Across six locally served models under 4B parameters, three data sources, and four feedback modes, larger registries sharply reduce single-pass TAAG pass rates across all sources. Repeated feedback improves some grounding metrics but costs substantially more tokens and does not consistently improve end-to-end F1; requirement-level TAAG-R feedback gives only modest gains over stage-only TAAG-S. The results suggest these evaluation signals help locate failures and probe corrections but do not by themselves solve registry scaling.

### Source excerpt

arXiv:2611.18622v1 Announce Type: cross Abstract: Tool-calling evaluations often report whether the final call matches a reference answer, but this single outcome merges different failures: choosing a function outside the available registry, violating the selected schema, or filling arguments with values not supported by the user request. These failures matter more when models must choose from larger tool registries. We introduce RFCD, a registry-scaled function-calling benchmark built from BFCL, Glaive Function Calling v2, and Tool-Call-Data, with standardized JSON-schema registries ranging from 5 to 3,000 functions. We also introduce TAAG, a deterministic stage-decomposed evaluator for registry conformance, structural completeness, and argument grounding. Across six locally served sub-4B models, three data sources, and four feedback modes, registry scale is the dominant source of degradation: single-pass TAAG pass rate drops sharply from small to large registries on all sources. Repeated feedback improves some grounding metrics but adds substantial token cost and does not consistently improve end-to-end F1. Requirement-level TAAG-R feedback gives modest gains over stage-only TAAG-S feedback, which indicates that TAAG signals are useful for analysis and correction probes but do not solve registry scaling by themselves. RFCD and TAAG provide an evaluation resource for measuring where tool-calling systems fail as registries grow.

## The agent builds the app, the user claims the database: how Ideavo does it

DevFeed: [The agent builds the app, the user claims the database: how Ideavo does it](<https://devfeed.tech/articles/the-agent-builds-the-app-the-user-claims-the-database-how-ideavo-does-it-136085.md>)

Original publisher: [Read original article](<https://neon.com/blog/the-agent-builds-the-app-the-user-claims-the-database-how-ideavo-does-it>)

Author: Carlota Soto

Published: 2026-11-16T12:00:00Z

Content type: article

Language: en

Sources: [Neon Docs](<https://devfeed.tech/sources/blog-neon-docs.md>)

Topics: [AI Chat](<https://devfeed.tech/topics/ai-chat.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>), [self-hosted databases vs managed services](<https://devfeed.tech/topics/self-hosted-databases-vs-managed-services.md>), [self-service](<https://devfeed.tech/topics/self-service.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-app-builder](<https://devfeed.tech/tags/ai-app-builder.md>), [anthropic](<https://devfeed.tech/tags/anthropic.md>), [api](<https://devfeed.tech/tags/api.md>), [api-key](<https://devfeed.tech/tags/api-key.md>), [app](<https://devfeed.tech/tags/app.md>), [app-builder](<https://devfeed.tech/tags/app-builder.md>), [apps](<https://devfeed.tech/tags/apps.md>), [builder](<https://devfeed.tech/tags/builder.md>), [builds](<https://devfeed.tech/tags/builds.md>), [case-studies](<https://devfeed.tech/tags/case-studies.md>), [charging](<https://devfeed.tech/tags/charging.md>), [connect](<https://devfeed.tech/tags/connect.md>), [database](<https://devfeed.tech/tags/database.md>), [free](<https://devfeed.tech/tags/free.md>), [generate](<https://devfeed.tech/tags/generate.md>), [google](<https://devfeed.tech/tags/google.md>), [instead](<https://devfeed.tech/tags/instead.md>), [key](<https://devfeed.tech/tags/key.md>), [user](<https://devfeed.tech/tags/user.md>)

### AI overview

Ideavo's agent provisions a separate Neon Postgres 17 project for each app that needs a database, then writes the first connection string to the app's .env as DATABASE_URL. The credential stays out of the model context and chat history; the agent configures Drizzle, writes the schema, and runs the first migration. Users can also add a database through the database panel or Better Auth integration. By default, Ideavo manages the database, but users can claim it through a Neon API transfer link; the connection string remains unchanged, so the app continues using the same database. Ideavo chose this setup to keep generated apps portable and avoid operating a fleet of self-hosted databases. It considered Supabase, whose bundled auth and storage services did not fit its desired separation of concerns. Each app maps to one Neon project, region, and default branch.

### Source excerpt

Ideavo is a free AI app builder where users generate apps from natural language. Instead of charging a subscription, Ideavo has no subscription and no credits - users connect their own API key or AI subscription from OpenAI, Anthropic, Google, OpenRouter, or another provider, and the Ideavo agent builds with it.

## BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructions

DevFeed: [BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructions](<https://devfeed.tech/articles/bigcodebench-benchmarking-code-generation-with-diverse-function-calls-and-complex-instructions-134942.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2406.15877>)

Author: Terry Yue Zhuo, Minh Chien Vu, Jenny Chim, Han Hu, Wenhao Yu, Ratnadira Widyasari, Imam Nur Bani Yusuf, Haolan Zhan, Junda He, Indraneil Paul, Simon Brunner, Chen Gong, Thong Hoang, Armel Randy Zebaze

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Software Engineering](<https://devfeed.tech/sources/arxiv-software-engineering.md>)

Topics: [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>)

Tags: [benchmarking](<https://devfeed.tech/tags/benchmarking.md>), [code-generation](<https://devfeed.tech/tags/code-generation.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-se](<https://devfeed.tech/tags/cs-se.md>), [function](<https://devfeed.tech/tags/function.md>), [instructions](<https://devfeed.tech/tags/instructions.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [python](<https://devfeed.tech/tags/python.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>)

### AI overview

BigCodeBench evaluates whether language models can solve practical coding tasks by composing function calls and following complex instructions. It contains 1,140 tasks across 7 domains and 139 libraries; each task has an average of 5.6 test cases, with reported average branch coverage of 99%. BigCodeBench-Instruct provides a variant that converts docstrings into concise natural-language instructions. In evaluations of 60 models, the highest reported score was 60%, compared with 97% for humans, indicating a substantial gap on these tasks. These findings are limited to the benchmark and evaluation described in the supplied abstract.

### Source excerpt

arXiv:2406.15877v5 Announce Type: replace-cross Abstract: Task automation has been greatly empowered by the recent advances in Large Language Models (LLMs) via Python code, where the tasks ranging from software engineering development to general-purpose reasoning. While current benchmarks have shown that LLMs can solve tasks using programs like human developers, the majority of their evaluations are limited to short and self-contained algorithmic tasks or standalone function calls. Solving challenging and practical tasks requires the capability of utilizing diverse function calls as tools to efficiently implement functionalities like data analysis and web development. In addition, using multiple tools to solve a task needs compositional reasoning by accurately understanding complex instructions. Fulfilling both of these characteristics can pose a great challenge for LLMs.To assess how well LLMs can solve challenging and practical tasks via programs, we introduce BigCodeBench, a benchmark that challenges LLMs to invoke multiple function calls as tools from 139 libraries and 7 domains for 1,140 fine-grained tasks. To evaluate LLMs rigorously, each task encompasses 5.6 test cases with an average branch coverage of 99%. In addition, we propose a natural-language-oriented variant of BigCodeBench, BigCodeBench-Instruct, that automatically transforms the original docstrings into short instructions only with essential information. Our extensive evaluation of 60 LLMs shows that LLMs are not yet capable of following complex instructions to use function calls precisely, with scores up to 60%, significantly lower than the human performance of 97%. The results underscore the need for further advancements in this area.

## My Local AI Agent Needed a Bigger Brain, So I Gave It a Decentralized GPU

DevFeed: [My Local AI Agent Needed a Bigger Brain, So I Gave It a Decentralized GPU](<https://devfeed.tech/articles/my-local-ai-agent-needed-a-bigger-brain-so-i-gave-it-a-decentralized-gpu-134162.md>)

Original publisher: [Read original article](<https://hackernoon.com/my-local-ai-agent-needed-a-bigger-brain-so-i-gave-it-a-decentralized-gpu?source=rss>)

Author: zkzk gamal

Published: 2026-11-14T18:00:02Z

Content type: article

Language: en

Sources: [HackerNoon](<https://devfeed.tech/sources/hackernoon.md>)

Topics: [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [Local AI](<https://devfeed.tech/topics/local-ai.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [Ollama](<https://devfeed.tech/topics/ollama.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [api](<https://devfeed.tech/tags/api.md>), [app](<https://devfeed.tech/tags/app.md>), [ast](<https://devfeed.tech/tags/ast.md>), [backend](<https://devfeed.tech/tags/backend.md>), [calculator](<https://devfeed.tech/tags/calculator.md>), [decentralize-ai-hackathon](<https://devfeed.tech/tags/decentralize-ai-hackathon.md>), [decentralized](<https://devfeed.tech/tags/decentralized.md>), [decentralized-gpu](<https://devfeed.tech/tags/decentralized-gpu.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [hackernoon-top-story](<https://devfeed.tech/tags/hackernoon-top-story.md>), [local](<https://devfeed.tech/tags/local.md>), [local-ai](<https://devfeed.tech/tags/local-ai.md>), [local-llm](<https://devfeed.tech/tags/local-llm.md>), [nosana](<https://devfeed.tech/tags/nosana.md>), [zkzkagent](<https://devfeed.tech/tags/zkzkagent.md>)

### AI overview

The article describes adding Nosana GPU inference to zkzkAgent while keeping routing, tools, files and confirmations on the user's machine. The agent can use local Ollama by default or send model calls to Nosana through LangChain's ChatOpenAI client; a live-model lookup selects a currently served model when none is configured. In a reported trial, a 27B Nosana model planned and built a FastAPI calculator app through 10 sequential local tool calls after plan approval; the app loaded and its API responded successfully. The author also reports safe expression parsing with Python's ast module and input limits. The arrangement lets sensitive work stay local while larger-model tasks use Nosana, but model calls do leave the machine in that mode. The article identifies remaining risks: shell commands can run without confirmation, and plan-level confirmations need strengthening. Benchmarks and automatic fallback are proposed future work, not reported results.

### Source excerpt

Running everything locally has a ceiling, and I kept hitting it. Here's how I fixed it.

## Siri AI Developer Preview Lets iOS, iPadOS, and macOS Apps Integrate with Siri

DevFeed: [Siri AI Developer Preview Lets iOS, iPadOS, and macOS Apps Integrate with Siri](<https://devfeed.tech/articles/siri-ai-developer-preview-lets-ios-ipados-and-macos-apps-integrate-with-siri-132514.md>)

Original publisher: [Read original article](<https://www.infoq.com/news/2026/11/apple-siri-ai-integration/>)

Author: Sergio De Simone

Published: 2026-11-13T16:00:00Z

Content type: news

Language: en

Sources: [InfoQ](<https://devfeed.tech/sources/infoq.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [iOS](<https://devfeed.tech/topics/ios.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>), [ipad](<https://devfeed.tech/topics/ipad.md>), [macOS](<https://devfeed.tech/topics/macos.md>), [Google Play](<https://devfeed.tech/topics/google-play.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-developer](<https://devfeed.tech/tags/ai-developer.md>), [apple](<https://devfeed.tech/tags/apple.md>), [apple-siri-ai-integration](<https://devfeed.tech/tags/apple-siri-ai-integration.md>), [apps](<https://devfeed.tech/tags/apps.md>), [developer](<https://devfeed.tech/tags/developer.md>), [developer-preview](<https://devfeed.tech/tags/developer-preview.md>), [development](<https://devfeed.tech/tags/development.md>), [integrate](<https://devfeed.tech/tags/integrate.md>), [ios](<https://devfeed.tech/tags/ios.md>), [ipados](<https://devfeed.tech/tags/ipados.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [macos](<https://devfeed.tech/tags/macos.md>), [mobile](<https://devfeed.tech/tags/mobile.md>), [news](<https://devfeed.tech/tags/news.md>), [preview](<https://devfeed.tech/tags/preview.md>), [siri](<https://devfeed.tech/tags/siri.md>), [siri-ai](<https://devfeed.tech/tags/siri-ai.md>), [visionos](<https://devfeed.tech/tags/visionos.md>), [watchos](<https://devfeed.tech/tags/watchos.md>)

### AI overview

Apple's Siri AI Developer Preview, included in the 27.3 beta releases, lets apps expose content and custom actions to Siri through App Intents. Developers can use Skills to combine several actions into one experience, including requests that span apps. Unspecified requests route to the foreground app, an app with an active Live Activity, or an app already mentioned in the conversation; background apps receive requests only when named. Preview access is limited to developer builds and internal TestFlight testers. The text says Siri AI is currently unavailable in the EU on iOS, iPadOS, and watchOS, and notes that Apple Intelligence features may have usage limits, with extended premium access planned.

### Source excerpt

Apple released the Siri AI Developer Preview in the 27.3 beta releases of its operating systems, enabling developers and apps to integrate with the assistant and hand off actions to it. During the Preview, access to Siri AI is restricted to developer builds and internal TestFlight testers. By Sergio De Simone

## Where Should the Code Go? Locating Foundation-Model Errors in Bitemporal Inventory Simulation

DevFeed: [Where Should the Code Go? Locating Foundation-Model Errors in Bitemporal Inventory Simulation](<https://devfeed.tech/articles/where-should-the-code-go-locating-foundation-model-errors-in-bitemporal-inventory-simulation-131694.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.14671>)

Author: Durga Sandeep Saluru, Lanchen Song, Aditya Kumaran

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>), [arXiv: Software Engineering](<https://devfeed.tech/sources/arxiv-software-engineering.md>)

Topics: [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>), [Parallel reasoning](<https://devfeed.tech/topics/parallel-reasoning.md>)

Tags: [budget](<https://devfeed.tech/tags/budget.md>), [call](<https://devfeed.tech/tags/call.md>), [canonical](<https://devfeed.tech/tags/canonical.md>), [code](<https://devfeed.tech/tags/code.md>), [continuation](<https://devfeed.tech/tags/continuation.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-se](<https://devfeed.tech/tags/cs-se.md>), [diagnostic](<https://devfeed.tech/tags/diagnostic.md>), [error](<https://devfeed.tech/tags/error.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [foundation-model](<https://devfeed.tech/tags/foundation-model.md>), [interface](<https://devfeed.tech/tags/interface.md>), [inventory](<https://devfeed.tech/tags/inventory.md>), [models](<https://devfeed.tech/tags/models.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [simulation](<https://devfeed.tech/tags/simulation.md>), [tool-calling](<https://devfeed.tech/tags/tool-calling.md>)

### AI overview

In an exactly solvable lost-sales inventory simulation, code placement addresses distinct errors: canonicalizing reports visible at a cutoff removes report-selection errors, while exact continuation removes rollout errors; neither fixes the other, and the dominant error varies by model. The study evaluates 36,672 independently rescored requests across twelve configurations of closed- and open-weight models, with a forced-answer-call control. Strong models reach ceiling on raw reports. More reasoning is not a dependable remedy: forced answer calls sharply reduce high-effort performance for GPT-5.4 mini and DeepSeek V4 Pro, while automatic choice lets them gain or hold; the provider's forced call disables reasoning for DeepSeek V4 Flash, and gpt-oss runs out of output budget at high effort. The results suggest tool-calling interfaces can confound reasoning evaluations for some model-provider pairs.

### Source excerpt

arXiv:2611.14671v1 Announce Type: new Abstract: Operational reports carry two clocks: when an event happened and when its report arrived; later reports can revise earlier ones. A foundation-model component that maintains the state supported by the reports visible at a cutoff, then rolls a supplied scenario forward, can be helped by exact code before it, after it or inside the call, or asked to reason harder. We ask which remedy removes which error, for which models, on an exactly solvable lost-sales inventory environment: 36,672 independently rescored requests over twelve configurations of closed and open-weight models, with a forced-answer-call interface as a control and protocols frozen before the calls. Weak models fail in separable ways. Canonical input removes report-selection error, exact continuation removes rollout error, neither repairs the other, and which error dominates differs by model; strong models are at ceiling on raw reports. Asking for more reasoning is no substitute and can depend on the interface: under a forced answer call high effort collapses GPT-5.4 mini and DeepSeek V4 Pro, which gain or hold under automatic choice; on its provider the forced call disables reasoning for DeepSeek V4 Flash; and at high effort gpt-oss exhausts its output budget. The contribution is a controlled component diagnostic, released with every request, and evidence that for some model and provider pairs the tool-calling interface confounds reasoning evaluation.

## SpaceMCP: Vision Language Model Spatial Reasoning via Model Context Protocol

DevFeed: [SpaceMCP: Vision Language Model Spatial Reasoning via Model Context Protocol](<https://devfeed.tech/articles/spacemcp-vision-language-model-spatial-reasoning-via-model-context-protocol-133024.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.15680>)

Author: Ameen Izhac, Nobuko Yoshida, Ronald Clark

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [vlm](<https://devfeed.tech/topics/vlm.md>), [Computer vision](<https://devfeed.tech/topics/computer-vision.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>)

Tags: [3d](<https://devfeed.tech/tags/3d.md>), [3d-reconstruction](<https://devfeed.tech/tags/3d-reconstruction.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [camera](<https://devfeed.tech/tags/camera.md>), [computer-vision](<https://devfeed.tech/tags/computer-vision.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [graph](<https://devfeed.tech/tags/graph.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [llm](<https://devfeed.tech/tags/llm.md>), [mcp](<https://devfeed.tech/tags/mcp.md>), [model-context-protocol](<https://devfeed.tech/tags/model-context-protocol.md>), [real-time](<https://devfeed.tech/tags/real-time.md>), [spatial-reasoning](<https://devfeed.tech/tags/spatial-reasoning.md>), [vision-language-model](<https://devfeed.tech/tags/vision-language-model.md>), [vision-language-models](<https://devfeed.tech/tags/vision-language-models.md>), [vlm](<https://devfeed.tech/tags/vlm.md>), [vlms](<https://devfeed.tech/tags/vlms.md>)

### AI overview

SpaceMCP gives a vision-language model structured function calls to query a real-time scene graph for distances, object positions, and camera properties. The graph combines feed-forward 3D reconstruction for metric geometry with open-vocabulary segmentation to locate objects; the VLM uses these primitives to address more complex spatial tasks. The abstract reports that without additional training, SpaceMCP achieves state-of-the-art results on VSI-Bench, MMSI-Bench, and VSTI-Bench, outperforming recent spatially post-trained VLMs and improving its base model by 15-22 points per benchmark.

### Source excerpt

arXiv:2611.15680v1 Announce Type: new Abstract: Despite impressive advances in large language models (LLM), vision-language models (VLM) still exhibit limited spatial reasoning capabilities. Recent progress in feed-forward 3D reconstruction has enabled precise, real-time geometric understanding from visual inputs. In this work, we propose SpaceMCP, a framework that leverages these advances to introduce an MCP-style paradigm for spatial reasoning. Our approach augments a VLM with structured function calls, allowing it to query a scene graph for spatial information such as distances, object positioning, and camera properties. The scene graph is constructed in real time using feed-forward 3D reconstruction to recover metric geometry, combined with open-vocabulary segmentation to identify and localize relevant objects within the scene. Paired with the VLM's strong reasoning capabilities, the VLM is able to combine this primitive information to solve more complex spatial tasks. Without any training, SpaceMCP achieves state-of-the-art results on VSI-Bench, MMSI-Bench and VSTI-Bench, outperforming recent spatially post-trained VLMs and improving its base VLM by 15-22 points per benchmark. We argue that, given the contrast between the relatively slow progress in native VLM spatial reasoning and the rapid advances in feed-forward 3D computer vision, this paradigm offers a promising path forward for enabling robust spatial reasoning.

## AURA 2.0: From User-Driven Responses to Model-Initiated Conversations in Streaming Video

DevFeed: [AURA 2.0: From User-Driven Responses to Model-Initiated Conversations in Streaming Video](<https://devfeed.tech/articles/aura-2-0-from-user-driven-responses-to-model-initiated-conversations-in-streaming-video-132306.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.17048>)

Author: Shuhan Li, Huankang Guan, Xintong Guo, Yang Bo, Yuyang He, Dunyuan Xu, Jingyan Chen, Fang Liu, Tsz Kin Wan, Jinpeng Chen, Zhenqi He, Yuxuan Hu, Wenbo Li, Han Gao, Qijun Li, Long Chen, Hongsheng Li, Ru

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [vlm](<https://devfeed.tech/topics/vlm.md>), [Conversational AI](<https://devfeed.tech/topics/conversational-ai.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>), [Real-time AI interactions](<https://devfeed.tech/topics/real-time-ai-interactions.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>), [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>)

Tags: [ai-assistants](<https://devfeed.tech/tags/ai-assistants.md>), [aura](<https://devfeed.tech/tags/aura.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [capabilities](<https://devfeed.tech/tags/capabilities.md>), [conversations](<https://devfeed.tech/tags/conversations.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [grounding](<https://devfeed.tech/tags/grounding.md>), [information](<https://devfeed.tech/tags/information.md>), [interaction](<https://devfeed.tech/tags/interaction.md>), [interactive](<https://devfeed.tech/tags/interactive.md>), [large-language-model](<https://devfeed.tech/tags/large-language-model.md>), [memory](<https://devfeed.tech/tags/memory.md>), [model-weights](<https://devfeed.tech/tags/model-weights.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [performance](<https://devfeed.tech/tags/performance.md>), [streaming](<https://devfeed.tech/tags/streaming.md>), [streaming-video](<https://devfeed.tech/tags/streaming-video.md>), [tool-calling](<https://devfeed.tech/tags/tool-calling.md>), [tools](<https://devfeed.tech/tags/tools.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

AURA 2.0 extends streaming video assistants beyond user-prompted replies: Model-Initiated Conversation (MIC) lets it start discussions about events in an evolving video, while Stream-Aware Tool Calling (SATC) uses visual and conversational context to call external tools and incorporate their results. Interactive Visual Grounding (IVG) handles user-marked objects and returns frame-aligned bounding boxes linking responses to regions. The system is trained on a dedicated streaming interaction dataset using a three-stage scheme, and Long-Horizon Interaction Memory (LHIM) supports continuity across extended streams. The authors report leading performance among open-source streaming models on public streaming benchmarks and positive targeted evaluations of the new capabilities; the abstract provides no comparative scores or further evaluation details. Weights and inference code are released.

### Source excerpt

arXiv:2611.17048v1 Announce Type: cross Abstract: Most AI assistants respond only when prompted, remaining silent even when events in the video stream offer meaningful opportunities for conversation. To move beyond purely user-driven interaction, we present AURA 2.0, a streaming video large language model that introduces three core upgrades to AURA. The key advance is Model-Initiated Conversation (MIC), which enables the model to identify topics worth discussing in the evolving visual stream and initiate conversations at appropriate moments without a user query. Stream-Aware Tool Calling (SATC) enables the model to invoke external tools using the current visual and conversational context and incorporate returned information into the ongoing interaction. Interactive Visual Grounding (IVG) supports visual grounding in both user inputs and model outputs. The model can identify objects marked by users on video frames and produce frame-aligned bounding boxes to link its responses to specific objects and regions. To develop these capabilities, we construct a dedicated streaming interaction dataset and adopt a three-stage training scheme to integrate the new behaviors with AURA's existing streaming interaction abilities. We further incorporate Long-Horizon Interaction Memory (LHIM) to maintain contextual continuity over extended video streams. Evaluations on public streaming benchmarks show that AURA 2.0 achieves leading performance among open-source streaming models, while targeted evaluations further validate its new interaction capabilities. We open-source the model weights and inference code.

## Laravel AI SDK 1.4 Lets Agents Recover From Tool Failures

DevFeed: [Laravel AI SDK 1.4 Lets Agents Recover From Tool Failures](<https://devfeed.tech/articles/laravel-ai-sdk-1-4-lets-agents-recover-from-tool-failures-133561.md>)

Original publisher: [Read original article](<https://laravel-news.com/laravel-ai-sdk-1-4>)

Author: Paul Redmond

Published: 2026-11-12T23:39:51Z

Content type: release

Language: en

Sources: [Laravel](<https://devfeed.tech/sources/laravel.md>)

Topics: [Laravel](<https://devfeed.tech/topics/laravel.md>), [SDK](<https://devfeed.tech/topics/sdk.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [Resilience](<https://devfeed.tech/topics/resilience.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>), [human in the loop agent](<https://devfeed.tech/topics/human-in-the-loop-agent.md>)

Tags: [1-4](<https://devfeed.tech/tags/1-4.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-sdk](<https://devfeed.tech/tags/ai-sdk.md>), [articles](<https://devfeed.tech/tags/articles.md>), [cost](<https://devfeed.tech/tags/cost.md>), [exceptions](<https://devfeed.tech/tags/exceptions.md>), [failures](<https://devfeed.tech/tags/failures.md>), [first](<https://devfeed.tech/tags/first.md>), [join](<https://devfeed.tech/tags/join.md>), [laravel](<https://devfeed.tech/tags/laravel.md>), [laravel-ai](<https://devfeed.tech/tags/laravel-ai.md>), [laravel-ai-releases](<https://devfeed.tech/tags/laravel-ai-releases.md>), [like](<https://devfeed.tech/tags/like.md>), [model](<https://devfeed.tech/tags/model.md>), [news](<https://devfeed.tech/tags/news.md>), [resume](<https://devfeed.tech/tags/resume.md>), [run](<https://devfeed.tech/tags/run.md>), [sdk](<https://devfeed.tech/tags/sdk.md>), [tests](<https://devfeed.tech/tags/tests.md>), [tool](<https://devfeed.tech/tags/tool.md>)

### AI overview

Laravel AI SDK v1.4.0 lets agent runs recover from tool exceptions: the model receives the error as a failed tool result and can explain it or try another tool. The exception is still reported and `ToolFailed` fires; event-listener exceptions can still fail the run. Agent fakes now execute approved tools when resuming paused approvals and store their results. The release adds raw provider HTTP responses to classification, embedding, transcription, image, and reranking responses; these are null after serialization and for fully cached embeddings. OpenAI billing-related 429 errors now raise `InsufficientCreditsException` without retry, while other 429s remain rate-limit errors; both remain failoverable. Stream provider errors use the broadcast name `error`, retaining the provider code in the payload. Conversation title calls now emit agent events, but failures fall back to the prompt without failing the turn. The excerpt also lists provider tool-call fixes and upgrade behavior changes for approval tests, exception handling, billing errors, and stream listeners.

### Source excerpt

Laravel AI SDK 1.4 sends tool exceptions back to the model instead of ending the run, and lets agent fakes resume tool approvals in tests. The post Laravel AI SDK 1.4 Lets Agents Recover From Tool Failures appeared first on Laravel News. Join the Laravel Newsletter to get Laravel articles like this directly in your inbox.

## Operational Ontology on Snowflake: From Business Meaning to Governed AI Action

DevFeed: [Operational Ontology on Snowflake: From Business Meaning to Governed AI Action](<https://devfeed.tech/articles/operational-ontology-on-snowflake-from-business-meaning-to-governed-ai-action-131048.md>)

Original publisher: [Read original article](<https://medium.com/snowflake/operational-ontology-on-snowflake-from-business-meaning-to-governed-ai-action-6f3ef270d3f2?source=rss----34b6daafc07---4>)

Author: Tianxia Jia

Published: 2026-11-12T21:36:38Z

Content type: article

Language: en

Sources: [Snowflake Builders](<https://devfeed.tech/sources/snowflake-builders.md>)

Topics: [AI Integration Strategies](<https://devfeed.tech/topics/ai-integration-strategies.md>), [knowledge-engineering](<https://devfeed.tech/topics/knowledge-engineering.md>), [business logic](<https://devfeed.tech/topics/business-logic.md>), [semantic-layer](<https://devfeed.tech/topics/semantic-layer.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Authorization](<https://devfeed.tech/topics/authorization.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>), [schema design](<https://devfeed.tech/topics/schema-design.md>), [AI-Ready Data](<https://devfeed.tech/topics/ai-ready-data.md>)

Tags: [acting](<https://devfeed.tech/tags/acting.md>), [actions](<https://devfeed.tech/tags/actions.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-reasoning](<https://devfeed.tech/tags/ai-reasoning.md>), [also](<https://devfeed.tech/tags/also.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [article](<https://devfeed.tech/tags/article.md>), [authorization](<https://devfeed.tech/tags/authorization.md>), [best](<https://devfeed.tech/tags/best.md>), [bottleneck](<https://devfeed.tech/tags/bottleneck.md>), [business](<https://devfeed.tech/tags/business.md>), [governed-ai](<https://devfeed.tech/tags/governed-ai.md>), [meaning](<https://devfeed.tech/tags/meaning.md>), [ontology](<https://devfeed.tech/tags/ontology.md>), [operational](<https://devfeed.tech/tags/operational.md>), [snowflake](<https://devfeed.tech/tags/snowflake.md>)

### AI overview

The proposed Snowflake Operational Ontology architecture links governed business definitions to live operational state, domain rules, decision services, AI reasoning, and controlled execution. Its lifecycle is Observe -> Understand -> Evaluate -> Decide -> Execute -> Verify: Cortex Agents investigate and explain options, while specialized services validate or simulate plans; an authorized human approves actions before procedures dispatch them to external systems. Versioned snapshots, isolated simulation results, idempotency, and observed outcomes support traceability and verification. The article stresses that a system's acknowledgement does not prove an action completed. Snowflake Business Ontology governs concepts and relationships but does not execute rules or actions. Standard tables are the default for history and events; hybrid tables can serve selected indexed transactional workloads, with regional availability and quotas, but do not support streams. Applications therefore need another supported event source or dispatcher for hybrid-table state changes. The examples are illustrative synthetic reference implementations, not production integrations.

### Source excerpt

A reference architecture for AI agents to reason over operational state, simulate decisions, execute governed actions, and verify real-world outcomes. From Enterprise Knowledge to Operational Decisions At 1:30 PM, packing station PB-04 goes offline in a distribution center. Orders begin accumulating, and a carrier scheduled to depart at 5:00 PM may leave before its shipments are ready. An AI agent connected to warehouse data can identify the failed station, trace affected orders, and explain the shipping risk. But the warehouse manager needs to decide what to do. Can qualified workers move to another station? Can orders be rerouted? Which recovery plan has the best chance of meeting the deadline without creating another bottleneck? Answering these questions requires current operational state, resource capabilities, business constraints, and a way to evaluate proposed changes. Acting on a decision also requires authorization and confirmation that the intended changes occurred. In the earlier Ontology on Snowflake series, I explored how ontology definitions, enterprise data, semantic models, and AI reasoning establish a shared understanding of business information. Snowflake now has a product of Business Ontology for governed business definiton (Public Preview soon). This article builds on that foundation by introducing a Snowflake Operational Ontology Reference Architecture. It connects governed business meaning to operational state, constraints, simulation, AI reasoning, and controlled actions. The goal is to establish a reusable pattern. A warehouse, supply chain network, and data center require different operational models, but they can share architectural contracts and a common decision lifecycle. 1. What Makes an Ontology Operational? An ontology defines the concepts, relationships, and semantics of a domain. In our warehouse example, the ontology defines Packing Station, Worker, Order, and Carrier Commitment, along with relationships such as workers operating s

## Introducing Gemini 4 Hydrogen

DevFeed: [Introducing Gemini 4 Hydrogen](<https://devfeed.tech/articles/introducing-gemini-4-hydrogen-131400.md>)

Original publisher: [Read original article](<https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-hydrogen/>)

Author: Koray Kavukcuoglu

Published: 2026-11-12T18:00:00Z

Content type: release

Language: en

Sources: [Gemini Models](<https://devfeed.tech/sources/gemini-models.md>)

Topics: [Frontier AI](<https://devfeed.tech/topics/frontier-ai.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [AI in frontend development](<https://devfeed.tech/topics/ai-in-frontend-development.md>), [google-antigravity](<https://devfeed.tech/topics/google-antigravity.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-workflows](<https://devfeed.tech/tags/agentic-workflows.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-studio](<https://devfeed.tech/tags/ai-studio.md>), [airtable](<https://devfeed.tech/tags/airtable.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [android](<https://devfeed.tech/tags/android.md>), [android-studio](<https://devfeed.tech/tags/android-studio.md>), [antigravity](<https://devfeed.tech/tags/antigravity.md>), [api](<https://devfeed.tech/tags/api.md>), [arena-ai](<https://devfeed.tech/tags/arena-ai.md>), [argon](<https://devfeed.tech/tags/argon.md>), [artificial-analysis](<https://devfeed.tech/tags/artificial-analysis.md>), [carbon](<https://devfeed.tech/tags/carbon.md>), [gemini](<https://devfeed.tech/tags/gemini.md>), [gemini-4](<https://devfeed.tech/tags/gemini-4.md>), [gemini-4-hydrogen](<https://devfeed.tech/tags/gemini-4-hydrogen.md>), [gemini-models](<https://devfeed.tech/tags/gemini-models.md>), [hydrogen](<https://devfeed.tech/tags/hydrogen.md>), [none](<https://devfeed.tech/tags/none.md>)

### AI overview

Gemini 4 Hydrogen is presented as a fast model for everyday coding and agentic work, with four thinking levels (low, medium, high by default, and xhigh) to vary reasoning depth, latency, and cost. Google says it exceeds Argon on the Artificial Analysis Intelligence Index at high (51 vs. 50), reaching 54 at xhigh, and is priced at $1.50 per million input tokens and $7.50 per million output tokens across its 1-million-token context window. In Antigravity, Orchestrator can switch between Carbon and Hydrogen; on Terminal-Bench 4.0 it reportedly approached Carbon's task success while using 24% less quota and finishing 19% faster. Hydrogen is also positioned as a standalone upgrade from 3.8 Flash: Google reports 2.5x faster task completion and Terminal-Bench 4.1 scores of 57% vs. 18.2%. The article describes developer, enterprise, Gemini app, and Search availability, alongside claimed safety improvements. These are the company's reported results and rollout details.

### Source excerpt

an image of carbon with the number 4 around it

## How to Stop an AI Agent Stuck in an Infinite Loop

DevFeed: [How to Stop an AI Agent Stuck in an Infinite Loop](<https://devfeed.tech/articles/how-to-stop-an-ai-agent-stuck-in-an-infinite-loop-129837.md>)

Original publisher: [Read original article](<https://openrouter.ai/blog/tutorials/ai-agent-infinite-loop/>)

Author: OpenRouter

Published: 2026-11-12T00:00:00Z

Content type: tutorial

Language: en

Sources: [OpenRouter](<https://devfeed.tech/sources/openrouter.md>)

Topics: [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agent-sdk](<https://devfeed.tech/tags/agent-sdk.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [api](<https://devfeed.tech/tags/api.md>), [api-key](<https://devfeed.tech/tags/api-key.md>), [arguments](<https://devfeed.tech/tags/arguments.md>), [billing](<https://devfeed.tech/tags/billing.md>), [budget](<https://devfeed.tech/tags/budget.md>), [bug](<https://devfeed.tech/tags/bug.md>), [cost](<https://devfeed.tech/tags/cost.md>), [costs](<https://devfeed.tech/tags/costs.md>), [how-to](<https://devfeed.tech/tags/how-to.md>), [infinite](<https://devfeed.tech/tags/infinite.md>), [model](<https://devfeed.tech/tags/model.md>), [retry](<https://devfeed.tech/tags/retry.md>), [stop](<https://devfeed.tech/tags/stop.md>), [tool](<https://devfeed.tech/tags/tool.md>)

### AI overview

The guide recommends four layers to stop runaway agent retries: enable doom-loop detection, normalize equivalent tool inputs with a loopKey, cap each run by steps and cost, and set an API-key credit limit. In tests across six models, detection reduced calls from 314 to 64 and spend from $0.29 to $0.05 across 18 runs, but exact-argument matching could be evaded by small input changes. Per-run caps cover novel retries; the key limit is enforced before provider routing and protects against bugs or many concurrent sessions. With BYOK, include_byok_in_limit must be enabled. A request whose worst-case input plus max_tokens exceeds the remaining limit fails without charge; in-flight requests can leave spend slightly above the cap. The guide also notes that step caps may allow pending tool calls and a final turn to run, so actual calls can exceed the configured step count.

### Source excerpt

An AI agent gets stuck in an infinite loop when a model keeps calling a tool that fails, usually with the same arguments, because nothing in the agent loop tells it to stop. Each retry resends the whole conversation, so every turn costs more than the last, and a loop left running overnight can cost thousands of dollars.

## BotPsy: On-Device Personal Assistant with a Sub-Billion-Parameter Agentic Language Model

DevFeed: [BotPsy: On-Device Personal Assistant with a Sub-Billion-Parameter Agentic Language Model](<https://devfeed.tech/articles/botpsy-on-device-personal-assistant-with-a-sub-billion-parameter-agentic-language-model-129791.md>)

Original publisher: [Read original article](<https://huggingface.co/blog/qvac/botpsy>)

Author: Milan Gritta; Jihye Back; JinUk; Cho; Ohjoon Kwon; Patrik Lambert; Mathias Buus; Amril Nurman

Published: 2026-11-11T05:01:16Z

Content type: release

Language: en

Sources: [Hugging Face](<https://devfeed.tech/sources/hugging-face-blog.md>)

Topics: [AI Chat](<https://devfeed.tech/topics/ai-chat.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>), [rlvr](<https://devfeed.tech/topics/rlvr.md>), [post-training](<https://devfeed.tech/topics/post-training.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [LLM security](<https://devfeed.tech/topics/llm-security.md>), [online privacy](<https://devfeed.tech/topics/online-privacy.md>), [quantization](<https://devfeed.tech/topics/quantization.md>), [llama.cpp](<https://devfeed.tech/topics/llama-cpp.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [costs](<https://devfeed.tech/tags/costs.md>), [device](<https://devfeed.tech/tags/device.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [on-device](<https://devfeed.tech/tags/on-device.md>), [privacy](<https://devfeed.tech/tags/privacy.md>)

### AI overview

BotPsy is a 0.8B-parameter, on-device tool-use assistant built on Qwen3.5-0.8B. In the authors' tests against four other models, it roughly doubled its base model's scores on several agent benchmarks and led the tested models on five of six overall; Qwen3.5-2B-Think scored higher on Claw-Eval and some search and special-case tasks. BotPsy also improved instruction following and reduced unsafe responses in automated red-teaming, including under prompt injection, though these results are benchmark-specific. It was trained through supervised fine-tuning, teacher distillation, and two reinforcement-learning stages using tool-use and general-domain tasks. The authors describe it as most suitable for low- to medium-difficulty tasks, primarily in English. Multi-step research remains difficult, and the model makes mistakes; it is not intended for high-stakes or safety-critical decisions or sensitive applications.

### Source excerpt

Frontier cloud LLMs carry high operational costs that can fluctuate unexpectedly, and relying on such third-party models introduces vulnerabilities, as providers can modify, deny, or deprecate access at any time. Furthermore, proprietary models offer limited customization for specific workflows, while posing risks to privacy, data control, and transparency.

## Agent Architecture Cost-Success Trade-offs: A Controlled Evaluation across Enterprise-Style SOPs

DevFeed: [Agent Architecture Cost-Success Trade-offs: A Controlled Evaluation across Enterprise-Style SOPs](<https://devfeed.tech/articles/agent-architecture-cost-success-trade-offs-a-controlled-evaluation-across-enterprise-style-sops-128213.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.12953>)

Author: Subhrangshu Nandi, Hitesh Arora, Yunfei Bai, Udita Patel, Chirag Shah, Hannah Kwon, Wei Zhang, Weian Chen, Chloe Ruoqi Li, Alberto Sconfienza, Ayush Goyal, Yasmine Lee, Ming Xiong, Prathap Ramachandra

Published: 2026-11-11T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>), [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [agentic workflows](<https://devfeed.tech/topics/agentic-workflows.md>), [AI Architecture](<https://devfeed.tech/topics/ai-architecture.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>)

Tags: [1-4](<https://devfeed.tech/tags/1-4.md>), [2](<https://devfeed.tech/tags/2.md>), [3](<https://devfeed.tech/tags/3.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [added](<https://devfeed.tech/tags/added.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-architecture](<https://devfeed.tech/tags/agent-architecture.md>), [agent-orchestration](<https://devfeed.tech/tags/agent-orchestration.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agents](<https://devfeed.tech/tags/agents.md>), [aggregate](<https://devfeed.tech/tags/aggregate.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [architectures](<https://devfeed.tech/tags/architectures.md>), [arm](<https://devfeed.tech/tags/arm.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [style](<https://devfeed.tech/tags/style.md>), [trade](<https://devfeed.tech/tags/trade.md>)

### AI overview

In a controlled comparison across 13 enterprise SOPs, no agent architecture was a universal winner, and aggregate success did not significantly distinguish the designs. Success varied sharply by SOP--from under 10% on the hardest tasks to nearly 100% on the easiest--while added structure showed only a directional, not significant, potential benefit on complex branching workflows. Coordinated and skill-augmented arms used 1.4-4.7 times as many tokens and 2.6-5.8 times the latency as a single agent, with gains usually small and rarely significant, making the single agent the efficient default in these results. Skill quality mattered: high-quality skills raised success from 67.4% to 73.1%, while degraded skills performed worse. A targeted multi-agent redesign was the only arm reported to significantly exceed the baseline, reaching 75.6% versus 67.4%; the abstract therefore points to implementation quality as a key factor.

### Source excerpt

arXiv:2611.12953v1 Announce Type: new Abstract: Enterprises increasingly use large language model agents to execute standard operating procedures (SOPs): long, branching, tool-intensive workflows. Yet there is little controlled evidence for choosing among competing agentic architectures. Holding the model, tasks, and tools fixed and varying only the architecture, we compare four designs (a function-calling single agent, a skill-augmented agent, centralized multi-agent orchestration, and a triage swarm; eight arms) across 13 enterprise SOPs, and introduce a skill-definition quality rubric to build matched high and low-quality skill corpora. Three takeaways emerge. (1)~\textbf{There is no universal best architecture; the right choice is SOP-dependent.} On a given SOP, task success ranges from under $10\%$ on the hardest to near $100\%$ on the easiest; no architecture is statistically separable in aggregate, and added structure only plausibly pays off on complex, branching SOPs (a directional, not yet significant, trend). (2)~\textbf{Added architecture cost must be justified by a task-success gain.} Every coordinated arm (multi-agent or skill-augmented) spends $1.4$--$4.7\times$ the tokens and $2.6$--$5.8\times$ the latency of the single agent for gains that are usually small and rarely significant, making the single agent the efficient default; high-quality skills do lift success (to $73.1\%$ from $67.4\%$) while degraded skills fall below it. (3)~\textbf{Implementation quality outweighs the single- versus multi-agent choice.} A plausible multi-agent implementation loses to the single agent, yet a targeted redesign of it becomes the only arm to significantly beat the baseline ($75.6\%$ vs.\ $67.4\%$).

[Next page](<https://devfeed.tech/topics/function-calling.md?cursor=WyIyMDI2LTExLTExVDA1OjAwOjAwKzAwOjAwIiwgImVlMWZjZjRjLTNiNTctNDQ5OC04NWEwLTc2MjZhZjZlNjhlZSJd>)