# DeepSeek Harness

DeepSeek Harness is an open-source, plugin-based agent harness for composing and running coding agents, with configurable tools, skills, sandboxes, storage, scheduling, and runtime modes.

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

## ElephantAgent: Contextual State Continuity in Agentic Systems

DevFeed: [ElephantAgent: Contextual State Continuity in Agentic Systems](<https://devfeed.tech/articles/elephantagent-contextual-state-continuity-in-agentic-systems-142538.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2607.01919>)

Author: Jiankai Jin, Xiangzheng Zhang, Zhao Liu, Wenzhuo Xu, Dongdong Yang, Deyue Zhang, Quanchen Zou

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

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: [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [ai security](<https://devfeed.tech/topics/ai-security.md>), [LLM security](<https://devfeed.tech/topics/llm-security.md>), [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [Agentic AI](<https://devfeed.tech/topics/what-is-agentic-ai.md>), [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [agentic-coding](<https://devfeed.tech/topics/agentic-coding.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-systems](<https://devfeed.tech/tags/agentic-systems.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [attack](<https://devfeed.tech/tags/attack.md>), [context](<https://devfeed.tech/tags/context.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [deepseek-harness](<https://devfeed.tech/tags/deepseek-harness.md>), [digest](<https://devfeed.tech/tags/digest.md>), [execution](<https://devfeed.tech/tags/execution.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [memory](<https://devfeed.tech/tags/memory.md>), [out-of-band](<https://devfeed.tech/tags/out-of-band.md>), [overhead](<https://devfeed.tech/tags/overhead.md>), [planning](<https://devfeed.tech/tags/planning.md>), [prompt](<https://devfeed.tech/tags/prompt.md>), [prompt-injection](<https://devfeed.tech/tags/prompt-injection.md>), [protocol](<https://devfeed.tech/tags/protocol.md>), [rollback](<https://devfeed.tech/tags/rollback.md>), [security](<https://devfeed.tech/tags/security.md>), [state](<https://devfeed.tech/tags/state.md>), [system](<https://devfeed.tech/tags/system.md>)

### AI overview

ElephantAgent aims to prevent out-of-band tool and memory poisoning by checking whether a bounded, security-critical portion of an agent's context matches its latest authorized state. Before each planning and execution cycle, it recomputes a digest and verifies it against a linearizable ledger of state transitions maintained over replicated trusted hardware; historical traceability makes transitions attributable and recoverable. In tests, ElephantDSH, an integration with DeepSeek Harness, withstood the reported out-of-band attacks and recovered from in-band prompt injection in all trials. Of five evaluated defenses that check only incoming context, all leaked targeted content under rollback and three did so under rug-pull attacks. Normal coding tasks incurred 1.02-1.07x relative overhead over DSH, with model latency variation removed.

### Source excerpt

arXiv:2607.01919v2 Announce Type: replace Abstract: Agentic systems enhance their capabilities by invoking external tools and maintaining persistent memory. However, recent tool and memory poisoning attacks can covertly bias agent behavior. We observe that such attacks mounted out of band share a single enabler: each succeeds only by presenting the agent with a contextual state (e.g., tool state and memory) that is not the latest faithfully evolved one. This observation suggests that a continuity check over the contextual state covers this family of attacks. We present ElephantAgent, a protocol that enforces Contextual State Continuity over the bounded, security-critical subset of the agent's context. Before each planning-and-execution cycle, ElephantAgent recomputes the digest of the local contextual state and verifies it against the latest authorized digest in a linearizable ledger of state transitions, maintained over replicated trusted hardware. The protocol also provides Historical Traceability, a secondary guarantee that keeps state transitions attributable and recoverable. We integrate ElephantAgent into DeepSeek Harness (DSH) and implement ElephantDSH, an agentic system protected by contextual state verification. ElephantDSH withstood all out-of-band attack attempts, from rug pulls and rollbacks to attacks on its own mechanism, and recovered from an in-band prompt injection in all trials. By comparison, existing defenses typically check only what enters the context: of the five we evaluated, all leaked the targeted content under the rollback, and three also under the rug pull. ElephantDSH completed all normal coding tasks with a relative overhead of 1.02-1.07$\times$ over the original DSH with model latency variation removed.

## Readie: Low-Latency Code Sandboxes for AI Agents

DevFeed: [Readie: Low-Latency Code Sandboxes for AI Agents](<https://devfeed.tech/articles/readie-low-latency-code-sandboxes-for-ai-agents-139599.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.22171>)

Author: Arjun Sivaraman, Zhaoheng Li, Alan Shi, Sanjay Malisetty, Eugene Wu, Yongjoo Park

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

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

Topics: [Docker Sandboxes](<https://devfeed.tech/topics/docker-sandboxes.md>), [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [algorithm](<https://devfeed.tech/tags/algorithm.md>), [approximation](<https://devfeed.tech/tags/approximation.md>), [artifacts](<https://devfeed.tech/tags/artifacts.md>), [budget](<https://devfeed.tech/tags/budget.md>), [code](<https://devfeed.tech/tags/code.md>), [code-execution](<https://devfeed.tech/tags/code-execution.md>), [construction](<https://devfeed.tech/tags/construction.md>), [covering](<https://devfeed.tech/tags/covering.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-db](<https://devfeed.tech/tags/cs-db.md>), [cs-se](<https://devfeed.tech/tags/cs-se.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [execution](<https://devfeed.tech/tags/execution.md>), [latency](<https://devfeed.tech/tags/latency.md>), [low-latency](<https://devfeed.tech/tags/low-latency.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [packages](<https://devfeed.tech/tags/packages.md>), [public](<https://devfeed.tech/tags/public.md>), [runtime](<https://devfeed.tech/tags/runtime.md>), [sandboxes](<https://devfeed.tech/tags/sandboxes.md>)

### AI overview

Readie targets slow startup in code sandboxes used by AI agents. Instead of relying on one compact snapshot to cover diverse generated code, it learns a set of small, quickly initialized snapshots, each covering part of the workload and collectively covering most of it. The set is constructed from a large training workload by formulating snapshot selection as an optimization problem and solving it with an approximation algorithm. On 15,000 public and LLM-synthesized scripts, the abstract reports 3-22x lower startup latency and 1.3-3.3x lower end-to-end runtime than existing approaches at the same storage budget.

### Source excerpt

arXiv:2611.22171v1 Announce Type: cross Abstract: From quantitative research to personal assistance, AI agents act by iteratively generating and executing code. To decouple the agent harness from code execution, many agents offload execution to dedicated systems called code sandboxes. However, existing code sandboxes suffer from slow startup. Whether they rely on custom snapshots or a single fat image, the root cause is the same: it is hard to build a compact, pre-warmed snapshot that can be shared across diverse agent-generated code while covering nearly all the artifacts (e.g., packages, datasets) it requires. In this work, we propose learned startup snapshots for low-latency code sandboxes. Our insight is that while it is hard for a single snapshot to be both compact and cover most workloads, a set of snapshots can achieve both. Each snapshot is small, initializes quickly, and covers a fraction of agent-generated code; together, the set covers most of it. We learn this set from a large training workload by formulating snapshot construction as an optimization problem and solving it with an approximation algorithm. Planned on 15K public and LLM-synthesized scripts, our system, Readie, reduces startup latency by 3-22$\times$ and end-to-end runtime by 1.3-3.3$\times$ compared to existing approaches at the same storage budget.

## Migrating to Koin Compiler with Koin AI Skills

DevFeed: [Migrating to Koin Compiler with Koin AI Skills](<https://devfeed.tech/articles/migrating-to-koin-compiler-with-koin-ai-skills-136662.md>)

Original publisher: [Read original article](<https://blog.insert-koin.io/migrating-to-koin-compiler-with-koin-ai-skills-fe704e2af3af?source=rss----925561f2ecdf---4>)

Author: Gabriel Bronzatti Moro

Published: 2026-11-16T13:01:04Z

Content type: tutorial

Language: en

Sources: [Koin developers](<https://devfeed.tech/sources/koin-developers-medium.md>)

Topics: [koin](<https://devfeed.tech/topics/koin.md>), [Agent Skills](<https://devfeed.tech/topics/agent-skills.md>), [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [Android](<https://devfeed.tech/topics/android.md>), [multiplatform](<https://devfeed.tech/topics/multiplatform.md>)

Tags: [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-assisted](<https://devfeed.tech/tags/ai-assisted.md>), [ai-skills](<https://devfeed.tech/tags/ai-skills.md>), [android](<https://devfeed.tech/tags/android.md>), [code-assistant](<https://devfeed.tech/tags/code-assistant.md>), [code-quality](<https://devfeed.tech/tags/code-quality.md>), [compile](<https://devfeed.tech/tags/compile.md>), [compiler](<https://devfeed.tech/tags/compiler.md>), [deepseek](<https://devfeed.tech/tags/deepseek.md>), [deepseek-harness](<https://devfeed.tech/tags/deepseek-harness.md>), [dependency-graph](<https://devfeed.tech/tags/dependency-graph.md>), [dsl](<https://devfeed.tech/tags/dsl.md>), [hybrid](<https://devfeed.tech/tags/hybrid.md>), [ios](<https://devfeed.tech/tags/ios.md>), [koin](<https://devfeed.tech/tags/koin.md>), [kotlin](<https://devfeed.tech/tags/kotlin.md>), [kotlin-multiplatform](<https://devfeed.tech/tags/kotlin-multiplatform.md>), [migrate](<https://devfeed.tech/tags/migrate.md>), [migration](<https://devfeed.tech/tags/migration.md>), [minutes](<https://devfeed.tech/tags/minutes.md>), [modules](<https://devfeed.tech/tags/modules.md>), [skills](<https://devfeed.tech/tags/skills.md>)

### AI overview

The author reports migrating StreamPlayer-App from a hybrid Koin setup to a fully compiler-based approach in three minutes using Koin Skills with DeepSeek Harness. The starting app used Koin 4.2.2 and Compiler Plugin 1.2.3 with compileSafety enabled, combining compile-time annotated modules, explicitly loaded DSL modules, and modules attached lazily to navigation destinations. The /di-migration skill produced a 131-step plan and applied changes incrementally, checking Gradle builds and testing affected features in an Android emulator after each step. The account presents this validation cycle as a way to catch issues early and keep the app functional during migration; it does not provide independent performance comparisons or detail the final code changes.

### Source excerpt

Hi friends! 👋 Today, we're going to explore the power of Koin Skills and how they can make real-world development tasks significantly faster and more efficient. Using the Koin Migration Skill, I was able to migrate the StreamPlayer-App from the traditional Koin DSL to the new Koin Compiler in just three minutes. The result? A clean migration, well-structured code, and an impressive turnaround time. In this article, I'll walk you through the entire process, step by step, showing how to use Koin Skills to automate migration tasks while keeping code quality and developer control at the center of the experience. Producing good software with AI -- Image generated by GeminiCurrent Koin Setup in StreamPlayer-App The app currently runs on Koin 4.2.2 with the Koin Compiler Plugin 1.2.3 compileSafety = true without KSP. Its dependency graph follows a hybrid approach: - Annotated modules are discovered at compile time. - Plain DSL modules are loaded explicitly. - A few modules are attached lazily to specific navigation destinations. The application has a single Koin instance, initialized through a shared commonMainfunction, while Android and iOS entry points provide their respective platform-specific dependencies. Let's migrate to Koin Compiler! 📈 Our goal is to migrate this hybrid setup to a fully compiler-based approach, leveraging the powerful AI-assisted tools provided by the Koin team: Koin Skills 🤖. These skills provide specialized workflows that help developers understand existing configurations, automate migration steps, and maintain code quality throughout the process. Let's see how much we can accomplish with them! Install Koin Skills I am using here DeepSeek Harness as a code assistant (provided by DeepSeek), and it was very easy to install the skills I just ask: Install https://github.com/InsertKoinIO/koin-skills on my machine (user level)Run the migration skill Basically we just need to run /di-migration skill in a fresh session. /di-migration migrate to Koin Compi

## DeepSeek's coding agent is too tidy for its own good -- and benchmarks are punishing it

DevFeed: [DeepSeek's coding agent is too tidy for its own good -- and benchmarks are punishing it](<https://devfeed.tech/articles/deepseek-s-coding-agent-is-too-tidy-for-its-own-good-and-benchmarks-are-punishing-it-132590.md>)

Original publisher: [Read original article](<https://thenewstack.io/deepseek-harness-cleanup-benchmark/>)

Author: Amanda Caswell

Published: 2026-11-13T17:56:30Z

Content type: news

Language: en

Sources: [The New Stack](<https://devfeed.tech/sources/the-new-stack.md>)

Topics: [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [hermes-agent](<https://devfeed.tech/topics/hermes-agent.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-engineering](<https://devfeed.tech/tags/ai-engineering.md>), [all](<https://devfeed.tech/tags/all.md>), [background](<https://devfeed.tech/tags/background.md>), [based](<https://devfeed.tech/tags/based.md>), [bash](<https://devfeed.tech/tags/bash.md>), [behavior](<https://devfeed.tech/tags/behavior.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [built-in](<https://devfeed.tech/tags/built-in.md>), [coding](<https://devfeed.tech/tags/coding.md>), [coding-agent](<https://devfeed.tech/tags/coding-agent.md>), [deepseek](<https://devfeed.tech/tags/deepseek.md>), [deepseek-harness](<https://devfeed.tech/tags/deepseek-harness.md>), [deepseek-v4-1-flash](<https://devfeed.tech/tags/deepseek-v4-1-flash.md>), [documentation](<https://devfeed.tech/tags/documentation.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [evaluations](<https://devfeed.tech/tags/evaluations.md>), [glm-5-3-flash](<https://devfeed.tech/tags/glm-5-3-flash.md>), [harness](<https://devfeed.tech/tags/harness.md>), [open-source](<https://devfeed.tech/tags/open-source.md>)

### AI overview

DeepSeek Harness's automatic shutdown of background processes hurt its Terminal-Bench results: three tasks required those processes to keep running, and it passed only 2 of 36 attempts with DeepSeek-V4.1-Flash and GLM-5.3-Flash, while OpenCode, Pi, and Hermes Agent passed all 36. The behavior reduced its scores by 1.5-1.7 percentage points and was sufficient to explain its 1.3-point gap from the leader in one evaluation. The article says this shutdown is a harness behavior; DeepSeek's local job registry is designed for jobs that end with the harness, while jobs that must survive restarts or cross processes need another backend. More broadly, keeping model, provider, and reasoning effort constant still produced large differences between harnesses: Mini-swe-agent solved 51.4% of tasks versus OpenClaw's 27.3%, and Pi cost $0.29 per solved task versus OpenCode's $3.35 in the initial evaluation. These findings suggest benchmark results depend on orchestration as well as the model.

### Source excerpt

DeepSeek's open-source coding agent is losing points on a major benchmark for a habit developers usually treat as good practice: The post DeepSeek's coding agent is too tidy for its own good -- and benchmarks are punishing it appeared first on The New Stack.

## If you still have six AI windows open, what got simpler? Here's what it looks like to ask one office manager what happened and what needs you.

DevFeed: [If you still have six AI windows open, what got simpler? Here's what it looks like to ask one office manager what happened and what needs you.](<https://devfeed.tech/articles/if-you-still-have-six-ai-windows-open-what-got-simpler-here-s-what-it-looks-like-to-ask-one-office-manager-what-happened-and-what-needs-you-132647.md>)

Original publisher: [Read original article](<https://natesnewsletter.substack.com/p/oddboro-manage-multiple-ai-agents-launch>)

Author: Nate

Published: 2026-11-13T14:03:50Z

Content type: article

Language: en

Sources: [Nate's Substack](<https://devfeed.tech/sources/nate-s-substack.md>)

Topics: [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [User interface design](<https://devfeed.tech/topics/ui-design.md>), [free software](<https://devfeed.tech/topics/free-software.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-tools](<https://devfeed.tech/tags/ai-tools.md>), [amazon](<https://devfeed.tech/tags/amazon.md>), [amazon-leo](<https://devfeed.tech/tags/amazon-leo.md>), [app](<https://devfeed.tech/tags/app.md>), [aws](<https://devfeed.tech/tags/aws.md>), [bugs](<https://devfeed.tech/tags/bugs.md>), [building-software](<https://devfeed.tech/tags/building-software.md>), [claude](<https://devfeed.tech/tags/claude.md>), [codex](<https://devfeed.tech/tags/codex.md>), [coding](<https://devfeed.tech/tags/coding.md>), [community](<https://devfeed.tech/tags/community.md>), [context](<https://devfeed.tech/tags/context.md>), [manager](<https://devfeed.tech/tags/manager.md>), [office](<https://devfeed.tech/tags/office.md>), [terminal](<https://devfeed.tech/tags/terminal.md>), [windows](<https://devfeed.tech/tags/windows.md>)

### AI overview

Oddboro is an open-source desktop app that gives multiple AI agents a shared workspace, context, and coordination point, so a user can ask one office manager what happened and what needs attention. Agents can receive work, access authorized tools and folders, hand work to each other, and prepare Gmail replies for the user to send. Shared personas let agents consult the user's decision preferences, while office rules such as read-only access clarify permissions. In a demo, the manager gathered requirements for a daily email report and proposed an agent for the work; deciding which messages need human judgment still depends on business context. An early user used one agent for a morning report, but had to maintain file copies sent into Oddboro, exposing an unresolved handoff. Gmail connections remain limited pending Google verification. The downloadable version uses the graphical office; headless operation and other interfaces are described as future goals.

### Source excerpt

More agents means more to keep track of. Oddboro shows your whole operation as a little office, which is a lot easier to follow than six terminal windows (and it's a lot cuter).

## DeepSeek's Super-Agent Moves the Attention Bottleneck From Your Feed to Your Agents

DevFeed: [DeepSeek's Super-Agent Moves the Attention Bottleneck From Your Feed to Your Agents](<https://devfeed.tech/articles/deepseek-s-super-agent-moves-the-attention-bottleneck-from-your-feed-to-your-agents-124209.md>)

Original publisher: [Read original article](<https://hackernoon.com/deepseeks-super-agent-moves-the-attention-bottleneck-from-your-feed-to-your-agents?source=rss>)

Author: Doomscroll

Published: 2026-11-09T18:58:18Z

Content type: article

Language: en

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

Topics: [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>), [supervisor agent pattern](<https://devfeed.tech/topics/supervisor-agent-pattern.md>), [agent orchestration](<https://devfeed.tech/topics/agent-orchestration.md>), [human in the loop agent](<https://devfeed.tech/topics/human-in-the-loop-agent.md>), [Loop Engineering](<https://devfeed.tech/topics/loop-engineering.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [attention](<https://devfeed.tech/tags/attention.md>), [attention-economy](<https://devfeed.tech/tags/attention-economy.md>), [capability](<https://devfeed.tech/tags/capability.md>), [deepseek](<https://devfeed.tech/tags/deepseek.md>), [deepseek-attention-economy](<https://devfeed.tech/tags/deepseek-attention-economy.md>), [deepseek-harness-engineering](<https://devfeed.tech/tags/deepseek-harness-engineering.md>), [deepseek-super-agent](<https://devfeed.tech/tags/deepseek-super-agent.md>), [hackernoon-top-story](<https://devfeed.tech/tags/hackernoon-top-story.md>), [harness-engineering](<https://devfeed.tech/tags/harness-engineering.md>)

### AI overview

DeepSeek argues that human attention, rather than model capability, limits parallel agent work: a person can actively track only three to five sessions. Its Super-agent prototype uses a long-lived Router to dispatch isolated Worker sessions, run tasks continuously and proactively, and surface work without requiring users to track each session. In simulated workloads it completed tasks in a median 12.4 minutes versus 20.4 minutes for a single-session agent; DeepSeek says it is a research prototype, not an official product, and the tests do not establish performance in real multi-user settings. The essay's link to the attention economy remains a possibility, not a demonstrated shift: consumer platforms and ad markets remain dominant. The source also stresses that open harnesses can grant models access to files, commands, and other real-world actions, making human approval boundaries such as "do not merge or change anything on your own" a key safety choice.

### Source excerpt

DeepSeek's Super-agent essay says human attention, not model capability, now limits multi-agent work. Is it an obituary for the attention economy?

## CertClaim: Towards Certified Model-Claim Verification for Coding-Agent Services

DevFeed: [CertClaim: Towards Certified Model-Claim Verification for Coding-Agent Services](<https://devfeed.tech/articles/certclaim-towards-certified-model-claim-verification-for-coding-agent-services-114059.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.01950>)

Author: Jiacheng Lu, Qixuan Yang, Tao Song, Daxin Jiang, Jing Li, Zuo Bai, Haibing Guan

Published: 2026-11-03T05:00:00Z

Content type: article

Language: en

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

Topics: [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [auditing](<https://devfeed.tech/tags/auditing.md>), [coding-agent](<https://devfeed.tech/tags/coding-agent.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [verification](<https://devfeed.tech/tags/verification.md>)

### AI overview

CertClaim addresses a weakness in auditing coding-agent services: existing methods assume the claimed model is served in one fixed configuration, so legitimate changes to sampling, reasoning mode, or system prompt can cause false accusations, while cheap model derivatives may go undetected. CertClaim includes authorized deployments in the claim and checks whether any one deployment explains all service evidence, calibrating multiple evidence channels and fusing them within each deployment. The paper analyzes false-accusation rates and detection conditions, and reports experiments across eight claims, two agent harnesses, and three task suites. The authors report validity across tested settings, better separation of substitutes than existing methods, and resistance to potential adaptive attacks.

### Source excerpt

arXiv:2611.01950v1 Announce Type: new Abstract: Coding-agent services are priced by the model they claim, making the verification of such model claims crucial for their users. In this paper, we reveal that existing model auditing methods rely on a latent assumption that the deployment is static, i.e., the claimed model is always served in one reference configuration. This assumption does not hold for coding-agent services, whose providers may legitimately change the sampling preset, the reasoning mode or the system prompt without changing the model. As a result, existing methods either falsely accuse honest providers or miss cheap derivatives of the claimed model. To address this problem, we propose a certified model-claim verification method (dubbed CertClaim). Instead of testing the service against one reference deployment, we make the authorized deployments part of the claim and test whether some authorized deployment explains all the evidence of the service, by calibrating multiple evidence channels in every authorized deployment and fusing them within each deployment. We provide theoretical analyses of its false-accusation rate and detection condition. Extensive experiments on eight claims, two agent harnesses and three task suites verify the effectiveness of CertClaim, which is valid in every setting while separating substitutes far better than existing methods, as well as its resistance to potential adaptive attacks.

## DeepSeek open-sources Super-Agent, a coding agent that watches team chats for work

DevFeed: [DeepSeek open-sources Super-Agent, a coding agent that watches team chats for work](<https://devfeed.tech/articles/deepseek-open-sources-super-agent-a-coding-agent-that-watches-team-chats-for-work-111334.md>)

Original publisher: [Read original article](<https://www.helpnetsecurity.com/2026/11/02/deepseek-dsh-superagent/>)

Author: Anamarija Pogorelec

Published: 2026-11-02T07:09:55Z

Content type: news

Language: en

Sources: [Help Net Security](<https://devfeed.tech/sources/help-net-security.md>)

Topics: [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [allowed](<https://devfeed.tech/tags/allowed.md>), [chat](<https://devfeed.tech/tags/chat.md>), [coding](<https://devfeed.tech/tags/coding.md>), [coding-agent](<https://devfeed.tech/tags/coding-agent.md>), [cybersecurity](<https://devfeed.tech/tags/cybersecurity.md>), [deepseek](<https://devfeed.tech/tags/deepseek.md>), [github](<https://devfeed.tech/tags/github.md>), [github-repository](<https://devfeed.tech/tags/github-repository.md>), [merge](<https://devfeed.tech/tags/merge.md>), [news](<https://devfeed.tech/tags/news.md>), [open](<https://devfeed.tech/tags/open.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [programming](<https://devfeed.tech/tags/programming.md>), [prototype](<https://devfeed.tech/tags/prototype.md>), [research-prototype](<https://devfeed.tech/tags/research-prototype.md>), [scope](<https://devfeed.tech/tags/scope.md>), [security](<https://devfeed.tech/tags/security.md>), [software-development](<https://devfeed.tech/tags/software-development.md>), [sources](<https://devfeed.tech/tags/sources.md>), [team](<https://devfeed.tech/tags/team.md>), [watches](<https://devfeed.tech/tags/watches.md>), [work](<https://devfeed.tech/tags/work.md>)

### AI overview

DeepSeek's free GitHub research prototype, dsh-superagent, performs repository work only when an allowed user explicitly requests it or a team rule applies. Teams can broaden access to a chat or GitHub scope so anyone there can mention the agent. The host enforces access lists and rules; messages or quoted documents cannot expand permissions. In default mode, implicit requests are assessed and logged but not accepted. The source recommends reviewing those records before switching to auto mode. Auto-merge is controlled by a "Delivery requirements" prompt; the host adds no merge logic. By default, the agent does not enable auto-merge under its own account, so people merge unless the prompt specifies another account.

### Source excerpt

Deepseek released a research prototype of dsh-superagent, a coding agent that does work on a GitHub repository for you. Who can ask for work The agent gets to work in two situations: an allowed user asks for something, or a rule your team wrote covers it. You can also open up a chat or GitHub scope, and then anyone in it can request work just by mentioning the agent. A message, comment or quoted document ... More -> The post DeepSeek open-sources Super-Agent, a coding agent that watches team chats for work appeared first on Help Net Security.

## DeepSeek Open-Sources dsh-superagent, a Router-Worker Coding Agent

DevFeed: [DeepSeek Open-Sources dsh-superagent, a Router-Worker Coding Agent](<https://devfeed.tech/articles/deepseek-open-sources-dsh-superagent-a-router-worker-coding-agent-110624.md>)

Original publisher: [Read original article](<https://www.marktechpost.com/2026/11/01/deepseek-dsh-superagent-router-worker-coding-agent/>)

Author: Asif Razzaq

Published: 2026-11-01T07:47:17Z

Content type: news

Language: en

Sources: [MarkTechPost](<https://devfeed.tech/sources/marktechpost.md>)

Topics: [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>), [agentic-coding](<https://devfeed.tech/topics/agentic-coding.md>), [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [human in the loop agent](<https://devfeed.tech/topics/human-in-the-loop-agent.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-shorts](<https://devfeed.tech/tags/ai-shorts.md>), [applications](<https://devfeed.tech/tags/applications.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [asynchronous](<https://devfeed.tech/tags/asynchronous.md>), [attention](<https://devfeed.tech/tags/attention.md>), [authorization](<https://devfeed.tech/tags/authorization.md>), [autonomous](<https://devfeed.tech/tags/autonomous.md>), [coding](<https://devfeed.tech/tags/coding.md>), [coding-agent](<https://devfeed.tech/tags/coding-agent.md>), [deepseek](<https://devfeed.tech/tags/deepseek.md>), [editors-pick](<https://devfeed.tech/tags/editors-pick.md>), [for-devs](<https://devfeed.tech/tags/for-devs.md>), [github](<https://devfeed.tech/tags/github.md>), [harness](<https://devfeed.tech/tags/harness.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [mit](<https://devfeed.tech/tags/mit.md>), [new-releases](<https://devfeed.tech/tags/new-releases.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [parallel](<https://devfeed.tech/tags/parallel.md>), [prototype](<https://devfeed.tech/tags/prototype.md>), [python](<https://devfeed.tech/tags/python.md>), [research-prototype](<https://devfeed.tech/tags/research-prototype.md>), [router](<https://devfeed.tech/tags/router.md>), [sessions](<https://devfeed.tech/tags/sessions.md>), [simulation](<https://devfeed.tech/tags/simulation.md>), [software-engineering](<https://devfeed.tech/tags/software-engineering.md>), [staff](<https://devfeed.tech/tags/staff.md>), [tech-news](<https://devfeed.tech/tags/tech-news.md>), [technology](<https://devfeed.tech/tags/technology.md>), [worker](<https://devfeed.tech/tags/worker.md>)

### AI overview

DeepSeek's MIT-licensed dsh-superagent is a TypeScript research prototype that routes GitHub and optional Feishu requests into authorized Tickets, each handled in an isolated Worker session; results can return to the conversation, often as a pull request. A long-running Router handles interaction and dispatch, while a local panel surfaces work needing human attention. Proactive work defaults to `observe`, which records what would be accepted without acting; `auto` and `off` are also available. It runs locally in DeepSeek Harness and requires Node.js 22.19+ or 24+, git 2.32+, Python 3.8+, tmux, and GitHub CLI 2.48+. In DeepSeek's self-reported simulation of noisy synthetic messages and simulated tools, p95 completion was 15.7 minutes versus 29.5 minutes for a single-session baseline; that baseline did not call a subagent in five runs. Real coding quality was not measured, so the timing result does not establish performance on real tasks.

### Source excerpt

DeepSeek released dsh-superagent, an MIT-licensed research prototype that turns GitHub and Feishu requests into parallel worker sessions. In DeepSeek's own simulation, p95 task completion fell to 15.7 minutes from 29.5 minutes. The post DeepSeek Open-Sources dsh-superagent, a Router-Worker Coding Agent appeared first on MarkTechPost.

## Tracekit Box: Running AI Agents in Micro-VMs Where Every Way Out Is Decided and Signed

DevFeed: [Tracekit Box: Running AI Agents in Micro-VMs Where Every Way Out Is Decided and Signed](<https://devfeed.tech/articles/tracekit-box-running-ai-agents-in-micro-vms-where-every-way-out-is-decided-and-signed-108082.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.38458>)

Author: Bravish Ghosh (Cygnux Labs)

Published: 2026-10-30T04:00:00Z

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: [virtual machines](<https://devfeed.tech/topics/virtual-machines.md>), [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [Secret Scanning](<https://devfeed.tech/topics/secret-scanning.md>)

Tags: [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ce](<https://devfeed.tech/tags/cs-ce.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-os](<https://devfeed.tech/tags/cs-os.md>), [firecracker](<https://devfeed.tech/tags/firecracker.md>), [network](<https://devfeed.tech/tags/network.md>), [patch](<https://devfeed.tech/tags/patch.md>), [sandboxes](<https://devfeed.tech/tags/sandboxes.md>), [secrets](<https://devfeed.tech/tags/secrets.md>), [tracing](<https://devfeed.tech/tags/tracing.md>), [verify](<https://devfeed.tech/tags/verify.md>), [vm](<https://devfeed.tech/tags/vm.md>)

### AI overview

Tracekit Box runs each AI-agent job in a Firecracker micro-VM without a network card and routes all permitted actions through host-side doors over vsock. A signer the agent cannot access decides and records commands, outbound connections, secrets, model calls, and workspace patches. Host-side name resolution rejects private and cloud-metadata addresses; secrets remain outside the VM and are substituted by a proxy only in a header for bound hosts; returned changes are reviewed file by file. Each job fails closed and produces a .tkb bundle verifiable offline. The abstract reports 13 passing end-to-end checks on KVM, nine fuzzing findings fixed, and 102 verifiable signed records in a demo. Limitations: Firecracker's jailer is not yet used, the signer must be co-located, and internal processes are recorded only when crossing a door.

### Source excerpt

arXiv:2610.38458v1 Announce Type: cross Abstract: AI agents now install packages, call model APIs with their operators' keys and change code in repositories that deploy to production. The network, credentials and write access they need are also how they can leak data, steal a key or change a system nobody approved. Sandboxes contain an agent but leave little evidence of what crossed their boundary; tracing records what happened but contains nothing. We present Tracekit Box, an open-source system (MIT, v0) that does both. Each job runs in its own Firecracker micro-VM with no network card. The only ways out are doors on the host, served over vsock: commands, outbound connections, secrets, model calls and the workspace patch. At every door a Tracekit signer the agent cannot reach decides what passes and signs it into one run per box. Names are resolved on the host; private and cloud-metadata addresses are always refused; secrets never enter the VM, which sees placeholders the host proxy swaps for real values only in one header for bound hosts; and changes come back as a patch applied file by file after the signer decides each file. Box fails closed, and each job ends as a .tkb bundle anyone can verify offline. All 13 end-to-end checks pass on a real KVM runner (the test that runs them takes 28.5 s, including image pull and boot); 15 fuzz targets on the trust boundary produced nine findings, all fixed; and the demo run's 102 signed records verify. Box does not yet run Firecracker's jailer, needs a co-located signer, and records processes inside the VM only where they cross a door.

## fx is now available in T3 Code

DevFeed: [fx is now available in T3 Code](<https://devfeed.tech/articles/fx-is-now-available-in-t3-code-105312.md>)

Original publisher: [Read original article](<https://vercel.com/changelog/fx-is-now-available-in-t3-code>)

Author: Andrew Gadzik

Published: 2026-10-28T00:00:00Z

Content type: release

Language: en

Sources: [Vercel News](<https://devfeed.tech/sources/vercel-news.md>)

Topics: [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [control-plane](<https://devfeed.tech/topics/control-plane.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [agent-harness](<https://devfeed.tech/tags/agent-harness.md>), [agents](<https://devfeed.tech/tags/agents.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>), [control-plane](<https://devfeed.tech/tags/control-plane.md>), [desktop](<https://devfeed.tech/tags/desktop.md>), [fx](<https://devfeed.tech/tags/fx.md>), [mobile](<https://devfeed.tech/tags/mobile.md>), [models](<https://devfeed.tech/tags/models.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [providers](<https://devfeed.tech/tags/providers.md>), [sessions](<https://devfeed.tech/tags/sessions.md>), [skills](<https://devfeed.tech/tags/skills.md>), [web](<https://devfeed.tech/tags/web.md>)

### AI overview

fx can now be used as a provider in T3 Code on web, desktop, and mobile. Its built-in provider supports Supervised and Auto modes and the models and skills already used with fx. To get started, install fx, configure a model provider, download T3 Code, then add fx under Settings > Providers > Add provider.

### Source excerpt

You can now use fx, our open-source coding agent harness, in T3 Code on web, desktop, and mobile. T3 Code is an open-source control plane for coding agents, allowing you to add agents like fx as providers for your coding sessions. The built-in fx provider supports Supervised and Auto modes, along with the models and skills you already use with fx. Get started by installing fx and setting up your model provider. Then download T3 Code and add fx from Settings > Providers > Add provider. Read more

## HarnessMM: Recursive Self-Improvement for Harness Token Efficiency

DevFeed: [HarnessMM: Recursive Self-Improvement for Harness Token Efficiency](<https://devfeed.tech/articles/harnessmm-recursive-self-improvement-for-harness-token-efficiency-100049.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.32813>)

Author: Yuhua Jiang, Chang Xu, Feifei Gao, Li Zhao

Published: 2026-10-26T04:00:00Z

Content type: article

Language: en

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

Topics: [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>)

Tags: [ablation](<https://devfeed.tech/tags/ablation.md>), [agent-harnesses](<https://devfeed.tech/tags/agent-harnesses.md>), [arguments](<https://devfeed.tech/tags/arguments.md>), [callback](<https://devfeed.tech/tags/callback.md>), [codex](<https://devfeed.tech/tags/codex.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [deepseek-harness](<https://devfeed.tech/tags/deepseek-harness.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [feedback](<https://devfeed.tech/tags/feedback.md>), [graphs](<https://devfeed.tech/tags/graphs.md>), [harness](<https://devfeed.tech/tags/harness.md>), [humaneval](<https://devfeed.tech/tags/humaneval.md>), [interface](<https://devfeed.tech/tags/interface.md>), [mbpp](<https://devfeed.tech/tags/mbpp.md>), [model](<https://devfeed.tech/tags/model.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [recursive-self-improvement](<https://devfeed.tech/tags/recursive-self-improvement.md>), [reduce](<https://devfeed.tech/tags/reduce.md>), [rsi](<https://devfeed.tech/tags/rsi.md>), [token-efficiency](<https://devfeed.tech/tags/token-efficiency.md>)

### AI overview

HarnessMM reduces repeated reasoning callbacks by reusing prior decisions only when observed tool receipts satisfy stored public guards. Its executable decision graphs preserve exact actions, arguments, and causal dependencies; new decisions, unsupported evidence, or selector abstention return control to the original actor, whose weights remain fixed. Recursive self-improvement refines companion harness configurations from training feedback, with validation choosing the deployment configuration. On HumanEval, Codex success remains 90.82% as tokens per task fall 29.54%; for DeepSeek Harness, success increases from 71.43% to 85.71% while tokens per task fall 87.93%. An MBPP ablation reports 26.91% lower cost than single-use graphs and 19.52% lower cost than retained callbacks.

### Source excerpt

arXiv:2610.32813v1 Announce Type: new Abstract: Agent harnesses repeatedly invoke reasoning after tool feedback, even when an earlier call has already fixed the next action. We introduce HarnessMM (Harness Minus Minus), a method for improving harness token efficiency through guarded decision reuse and recursive self-improvement (RSI). RSI iteratively refines companion harness configurations from training feedback, with validation selecting the configuration for deployment. A controller at the model-tool boundary reuses resolved decisions to replace later reasoning callbacks while actor weights remain fixed. Executable Decision Graphs store exact actions, literal arguments, causal dependencies, and public guards. Guarded Decision Reuse activates these persistent decisions from observed receipts; Certified Callback Substitution executes them through the native interface and records removal of the registered slow callback. New semantic decisions, unsupported evidence, and selector abstention invoke the original actor again. The representation restricts what feedback may change, while complete-task measurements establish quality and cost. On HumanEval, HarnessMM preserves Codex success at 90.82\% while reducing Tokens/task (k) from 66.72 to 47.02 (29.54\% saving). For DeepSeek Harness, success rises from 71.43\% to 85.71\% while Tokens/task (k) fall from 96.28 to 11.62 (87.93\% saving). A separate MBPP ablation measures 26.91\% lower cost than single-use graphs and 19.52\% lower cost than retained callbacks. These results establish guarded decision reuse as a practical way to reduce repeated reasoning in existing harnesses.

## The five components of building a "company harness"

DevFeed: [The five components of building a "company harness"](<https://devfeed.tech/articles/the-five-components-of-building-a-company-harness-100224.md>)

Original publisher: [Read original article](<https://www.langchain.com/blog/components-building-company-harness>)

Author: LangChain Accounts

Published: 2026-10-25T18:06:32Z

Content type: article

Language: en

Sources: [LangChain Blog](<https://devfeed.tech/sources/langchain-blog.md>)

Topics: [LangChain](<https://devfeed.tech/topics/langchain.md>), [Slack](<https://devfeed.tech/topics/slack.md>), [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [building](<https://devfeed.tech/tags/building.md>), [company](<https://devfeed.tech/tags/company.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [harness](<https://devfeed.tech/tags/harness.md>), [langchain](<https://devfeed.tech/tags/langchain.md>), [marketing](<https://devfeed.tech/tags/marketing.md>), [sales](<https://devfeed.tech/tags/sales.md>), [teams](<https://devfeed.tech/tags/teams.md>)

### AI overview

LangChain's company harness combines five shared primitives: Slack-based access, identity-aware service actions, global and user-level memory, sandboxed code execution with authenticated service access, and scheduled or event-triggered runs. Its three agents apply these differently: OpenSWE uses the invoking user's GitHub identity for requests but its GitHub app identity for background reviews; GTM Agent uses fixed Salesforce credentials and user-authorized email available only in DMs; Content Studio uses user identity for Notion and company credentials for Twitter, with posting limited to whitelisted users. The agents run code in sandboxes connected through LangSmith's secret proxy. Slack supports direct messages, channel mentions, and breakout threads, while a custom UI handles larger email-review backlogs. The excerpt says memory can be global or user-specific and that this area remains underdeveloped. All three are built on Managed Deep Agents, which provides channels, crons, sandboxes, memory, and identity with room to customize behavior.

### Source excerpt

Learn how LangChain built internal agents across marketing, sales, and engineering, and how they have completely changed how those teams work.

## Rho said it was running. It wasn't.

DevFeed: [Rho said it was running. It wasn't.](<https://devfeed.tech/articles/rho-said-it-was-running-it-wasn-t-96065.md>)

Original publisher: [Read original article](<https://www.swytchcode.com/content/rho-said-it-was-running-it-wasn-t>)

Author: Swytchcode Team

Published: 2026-10-22T04:29:10Z

Content type: tutorial

Language: en

Sources: [Swytchcode](<https://devfeed.tech/sources/swytchcode.md>)

Topics: [Post Mortem](<https://devfeed.tech/topics/post-mortem.md>), [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [Tool](<https://devfeed.tech/topics/tool.md>), [runc](<https://devfeed.tech/topics/runc.md>), [Documentation](<https://devfeed.tech/topics/documentation.md>)

Tags: [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [debug](<https://devfeed.tech/tags/debug.md>), [extensions](<https://devfeed.tech/tags/extensions.md>), [how-to](<https://devfeed.tech/tags/how-to.md>), [pi](<https://devfeed.tech/tags/pi.md>), [rules](<https://devfeed.tech/tags/rules.md>), [shell](<https://devfeed.tech/tags/shell.md>)

### AI overview

A fresh Rho 0.1.12 install can report "running" after Pi has quit: Rho still installs the old Pi package, its status check only confirms launch, and the review module also lacks `glob`. The excerpt's workaround is to install the compatible MCP adapter version 2.5.4, disable `subagents` and `web-access`, install `glob`, and re-pin the adapter after `rho sync`. With Rho running, seven repeatable checks against mocked Stripe passed, including blocking writes pending human approval and refusing attempts to change Swytchcode rules. But Rho's unrestricted shell could edit the rules file directly, so the authors added a Pi extension that blocks `bash`, `interactive_shell`, `write`, and `edit`, and blocks reads from Swytchcode folders; a container with Swytchcode on the host over HTTP also passed the checks. Setup requires a Swytchcode account, and approvals require Business or Enterprise. Real models, Telegram/email, and phone approvals were not tested.

### Source excerpt

We connected Rho to Swytchcode, and a fresh install told us it was running while its agent had already quit. Here's what broke, how we fixed it, and the small Pi extension that keeps Rho's shell away from Swytchcode's rules.

## DeepSeek open-sources the inference engine behind its API. And Huawei gets equal billing with Nvidia.

DevFeed: [DeepSeek open-sources the inference engine behind its API. And Huawei gets equal billing with Nvidia.](<https://devfeed.tech/articles/deepseek-open-sources-the-inference-engine-behind-its-api-and-huawei-gets-equal-billing-with-nvidia-90665.md>)

Original publisher: [Read original article](<https://thenewstack.io/deepseek-deepinfer-huawei-nvidia/>)

Author: Paul Sawers

Published: 2026-10-20T14:58:36Z

Content type: news

Language: en

Sources: [The New Stack](<https://devfeed.tech/sources/the-new-stack.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [AWS Trainium](<https://devfeed.tech/topics/aws-trainium.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-chips](<https://devfeed.tech/tags/ai-chips.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [apache-2-0](<https://devfeed.tech/tags/apache-2-0.md>), [asynchronous](<https://devfeed.tech/tags/asynchronous.md>), [batch](<https://devfeed.tech/tags/batch.md>), [chips](<https://devfeed.tech/tags/chips.md>), [code](<https://devfeed.tech/tags/code.md>), [community](<https://devfeed.tech/tags/community.md>), [compute-costs](<https://devfeed.tech/tags/compute-costs.md>), [computing](<https://devfeed.tech/tags/computing.md>), [deepseek](<https://devfeed.tech/tags/deepseek.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [huawei](<https://devfeed.tech/tags/huawei.md>), [inference](<https://devfeed.tech/tags/inference.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [stack](<https://devfeed.tech/tags/stack.md>)

### AI overview

DeepSeek has released DeepInfer, the inference engine it uses for its models, under the Apache 2.0 license. It supports offline inference and includes optimizations for distributing work across chips, speculative decoding, asynchronous scheduling, structured outputs, and request handling through the companion deepseek-recipe library. The release supports Nvidia SM100 GPUs and Huawei Ascend 950 NPUs, though some features are Nvidia-only. It does not include several components needed for a complete production service, such as KV cache management, cross-machine inference stages, and replica load balancing.

### Source excerpt

DeepSeek has open-sourced the inference engine it uses to serve its models in production, a step it said last year The post DeepSeek open-sources the inference engine behind its API. And Huawei gets equal billing with Nvidia. appeared first on The New Stack.

## DeepSeek Just Released V4.1-Flash-1017, Scores 77.0 on DeepSWE, Opens Search API

DevFeed: [DeepSeek Just Released V4.1-Flash-1017, Scores 77.0 on DeepSWE, Opens Search API](<https://devfeed.tech/articles/deepseek-just-released-v4-1-flash-1017-scores-77-0-on-deepswe-opens-search-api-89247.md>)

Original publisher: [Read original article](<https://www.marktechpost.com/2026/10/17/deepseek-just-released-v4-1-flash-1017-scores-77-0-on-deepswe-opens-search-api/>)

Author: Asif Razzaq

Published: 2026-10-17T08:53:58Z

Content type: news

Language: en

Sources: [MarkTechPost](<https://devfeed.tech/sources/marktechpost.md>)

Topics: [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [Chat completions endpoints](<https://devfeed.tech/topics/chat-completions-endpoints.md>), [Claude](<https://devfeed.tech/topics/claude.md>), [AI-driven information retrieval](<https://devfeed.tech/topics/ai-driven-information-retrieval.md>)

Tags: [agent-harness](<https://devfeed.tech/tags/agent-harness.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-paper-summary](<https://devfeed.tech/tags/ai-paper-summary.md>), [ai-shorts](<https://devfeed.tech/tags/ai-shorts.md>), [api](<https://devfeed.tech/tags/api.md>), [applications](<https://devfeed.tech/tags/applications.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [claude](<https://devfeed.tech/tags/claude.md>), [claude-opus-5](<https://devfeed.tech/tags/claude-opus-5.md>), [code](<https://devfeed.tech/tags/code.md>), [deepseek](<https://devfeed.tech/tags/deepseek.md>), [deepseek-v4-1-flash](<https://devfeed.tech/tags/deepseek-v4-1-flash.md>), [deepseek-v4-1-flash-1017](<https://devfeed.tech/tags/deepseek-v4-1-flash-1017.md>), [deepswe-v1-1](<https://devfeed.tech/tags/deepswe-v1-1.md>), [editors-pick](<https://devfeed.tech/tags/editors-pick.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [large-language-model](<https://devfeed.tech/tags/large-language-model.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [new-releases](<https://devfeed.tech/tags/new-releases.md>), [release](<https://devfeed.tech/tags/release.md>), [search](<https://devfeed.tech/tags/search.md>), [search-api](<https://devfeed.tech/tags/search-api.md>), [staff](<https://devfeed.tech/tags/staff.md>), [tech-news](<https://devfeed.tech/tags/tech-news.md>), [technology](<https://devfeed.tech/tags/technology.md>)

### AI overview

DeepSeek released V4.1-Flash-1017, a post-training checkpoint aimed at coding and agent workflows, and opened its Search and Fetch APIs to developers. The article reports DeepSeek's benchmark results, model architecture and pricing, along with API endpoints and usage limits. It notes that the model leads on some reported benchmarks while Claude Opus 5 leads on others.

### Source excerpt

DeepSeek has just released DeepSeek-V4.1-Flash-1017, a new checkpoint of its V4.1-Flash model. It is live on the DeepSeek website, app and API. DeepSeek claims stronger scores, faster task completion and shorter outputs than the September V4.1-Flash release. The company calls it a Pareto improvement in performance and cost. Alongside it, DeepSeek opened its Search and [...] The post DeepSeek Just Released V4.1-Flash-1017, Scores 77.0 on DeepSWE, Opens Search API appeared first on MarkTechPost.

## Lakebase and Agentic SDLC: Branching Databases for Coding Agents

DevFeed: [Lakebase and Agentic SDLC: Branching Databases for Coding Agents](<https://devfeed.tech/articles/lakebase-and-agentic-sdlc-branching-databases-for-coding-agents-77639.md>)

Original publisher: [Read original article](<https://www.databricks.com/blog/lakebase-and-agentic-sdlc-branching-databases-coding-agents>)

Author: Thibaut Gourdel

Published: 2026-10-08T22:00:00Z

Content type: tutorial

Language: en

Sources: [Databricks](<https://devfeed.tech/sources/databricks.md>)

Topics: [Application Development](<https://devfeed.tech/topics/application-development.md>), [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [Continuous integration](<https://devfeed.tech/topics/continuous-integration.md>), [CI/CD](<https://devfeed.tech/topics/cicd.md>)

Tags: [cloudflare](<https://devfeed.tech/tags/cloudflare.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>), [continuous-integration](<https://devfeed.tech/tags/continuous-integration.md>), [databases](<https://devfeed.tech/tags/databases.md>), [lakebase](<https://devfeed.tech/tags/lakebase.md>), [platform](<https://devfeed.tech/tags/platform.md>), [product](<https://devfeed.tech/tags/product.md>), [sdlc](<https://devfeed.tech/tags/sdlc.md>)

### AI overview

The article demonstrates a Lakebase workflow that gives each coding agent and pull request an isolated Postgres database branch for schema changes and tests. It also describes applying migrations through CI, testing preview deployments, and using temporary branches to debug production-derived data safely.

### Source excerpt

AI has changed how software gets built. As coding agents take on a growing share...

## Revamping Skills in Deep Agents

DevFeed: [Revamping Skills in Deep Agents](<https://devfeed.tech/articles/revamping-skills-in-deep-agents-78353.md>)

Original publisher: [Read original article](<https://www.langchain.com/blog/revamping-skills-in-deep-agents>)

Author: LangChain Accounts

Published: 2026-10-08T11:05:30Z

Content type: article

Language: en

Sources: [LangChain Blog](<https://devfeed.tech/sources/langchain-blog.md>)

Topics: [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [Agent Skills](<https://devfeed.tech/topics/agent-skills.md>), [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [context](<https://devfeed.tech/tags/context.md>), [repositories](<https://devfeed.tech/tags/repositories.md>), [skills](<https://devfeed.tech/tags/skills.md>)

### AI overview

Deep Agents adds three ways to manage agent skills: bind tools so they load alongside the relevant skill, pin skills before a model call, and reload skills during a thread. The changes aim to keep context efficient while supporting large skill libraries and current instructions.

### Source excerpt

Deep Agents now lets you bind tools to skills, pin skills at runtime, and reload skills mid-thread, so agents with expansive skill repositories stay context-efficient and effective.

## The immortal life of Pi (Running the Pi coding agent on Temporal)

DevFeed: [The immortal life of Pi (Running the Pi coding agent on Temporal)](<https://devfeed.tech/articles/the-immortal-life-of-pi-running-the-pi-coding-agent-on-temporal-77370.md>)

Original publisher: [Read original article](<https://temporal.io/blog/the-immortal-life-of-pi-running-the-pi-coding-agent-on-temporal>)

Author: Moe Abadi

Published: 2026-10-08T00:00:00Z

Content type: article

Language: en

Sources: [Temporal](<https://devfeed.tech/sources/temporal-blog.md>)

Topics: [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>), [Service fabric](<https://devfeed.tech/topics/service-fabric.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [cursor](<https://devfeed.tech/topics/cursor.md>), [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [human in the loop agent](<https://devfeed.tech/topics/human-in-the-loop-agent.md>), [Submodules](<https://devfeed.tech/topics/submodules.md>), [Cloudflare Workers](<https://devfeed.tech/topics/cloudflare-workers.md>)

Tags: [agent-harness](<https://devfeed.tech/tags/agent-harness.md>), [agent-harnesses](<https://devfeed.tech/tags/agent-harnesses.md>), [checkpoint](<https://devfeed.tech/tags/checkpoint.md>), [clock](<https://devfeed.tech/tags/clock.md>), [crash](<https://devfeed.tech/tags/crash.md>), [directory](<https://devfeed.tech/tags/directory.md>), [distributed](<https://devfeed.tech/tags/distributed.md>), [docker](<https://devfeed.tech/tags/docker.md>), [durable-objects](<https://devfeed.tech/tags/durable-objects.md>), [earendil](<https://devfeed.tech/tags/earendil.md>), [executor](<https://devfeed.tech/tags/executor.md>), [fail](<https://devfeed.tech/tags/fail.md>), [failed](<https://devfeed.tech/tags/failed.md>), [failover](<https://devfeed.tech/tags/failover.md>), [failure](<https://devfeed.tech/tags/failure.md>), [files](<https://devfeed.tech/tags/files.md>), [pi](<https://devfeed.tech/tags/pi.md>), [temporal-concepts](<https://devfeed.tech/tags/temporal-concepts.md>)

### AI overview

The article describes pi-temporal, a retrofit that runs the Pi coding agent on Temporal Workers and lets another Worker resume a turn after a machine failure. It explains how the session transcript and Git bundles preserve conversation and project state, and how claims, leases, and outcome-unknown reports prevent automatic re-execution of interrupted tool calls. It also discusses limitations, including uncertain external effects and the infrastructure needed for recovery.

### Source excerpt

When a machine dies mid-command, the Pi coding agent keeps going. See how pi-temporal uses Temporal Workers and Durable Execution to recover agent turns.

## Arcade vs Composio vs Nango vs Swytchcode (2026): Which Agent Tool Layer Fits?

DevFeed: [Arcade vs Composio vs Nango vs Swytchcode (2026): Which Agent Tool Layer Fits?](<https://devfeed.tech/articles/arcade-vs-composio-vs-nango-vs-swytchcode-2026-which-agent-tool-layer-fits-90687.md>)

Original publisher: [Read original article](<https://www.swytchcode.com/content/arcade-vs-composio-vs-nango-vs-swytchcode>)

Author: Swytchcode Team

Published: 2026-10-06T11:00:00Z

Content type: comparison

Language: en

Sources: [Swytchcode](<https://devfeed.tech/sources/swytchcode.md>)

Topics: [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>), [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [agent](<https://devfeed.tech/tags/agent.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [audit](<https://devfeed.tech/tags/audit.md>), [comparison](<https://devfeed.tech/tags/comparison.md>), [idempotency](<https://devfeed.tech/tags/idempotency.md>), [policy](<https://devfeed.tech/tags/policy.md>), [pricing](<https://devfeed.tech/tags/pricing.md>)

### AI overview

This article compares Arcade, Composio, Nango, and Swytchcode as layers for connecting AI agents to external APIs. It explains differences in user authorization, hosted tool catalogs, code-owned integrations, API-call policy, approvals, retries, idempotency, audit logs, infrastructure, and pricing. The article positions the products for different needs and describes ways to combine them.

### Source excerpt

A criteria-based comparison of Arcade, Composio, Nango, and Swytchcode for AI agents in 2026: what each one does, whose credentials it uses, retries and idempotency, policy, audit, and published pricing.

## GPT Researcher drops embeddings for Jev

DevFeed: [GPT Researcher drops embeddings for Jev](<https://devfeed.tech/articles/gpt-researcher-drops-embeddings-for-jev-64944.md>)

Original publisher: [Read original article](<https://jevdaily.com/issues/2026-10-05-w7>)

Author: Jev Weekly

Published: 2026-10-05T17:00:00Z

Content type: news

Language: en

Sources: [Jev Daily](<https://devfeed.tech/sources/jev-daily.md>)

Topics: [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [open-responses](<https://devfeed.tech/topics/open-responses.md>), [Supabase](<https://devfeed.tech/topics/supabase.md>)

Tags: [also-on-the-timeline](<https://devfeed.tech/tags/also-on-the-timeline.md>), [built-with-it](<https://devfeed.tech/tags/built-with-it.md>), [someone-measured-it](<https://devfeed.tech/tags/someone-measured-it.md>), [the-case-against](<https://devfeed.tech/tags/the-case-against.md>)

### AI overview

GPT Researcher switched from embeddings to Jev by default after internal tests found relevant context in 73% of cases, compared with 46% for the embeddings path, at the same reported cost. The issue also covers model routers for coding agents, limitations reported for pg-jev, and Cloudflare's release of Jev API-compatible decision models. Direct public benchmarks comparing Jev with OpenAI's Decisions API were not reported.

### Source excerpt

Assaf Elovic measured 73% versus 46% relevant context swapping embeddings for Jev, and GPT Researcher now ships that way by default. The same week, pg-jev...

## DeepSeek Harness v0.2 Brings Official Desktop Apps to Its Open-Source Agent Harness

DevFeed: [DeepSeek Harness v0.2 Brings Official Desktop Apps to Its Open-Source Agent Harness](<https://devfeed.tech/articles/deepseek-harness-v0-2-brings-official-desktop-apps-to-its-open-source-agent-harness-64476.md>)

Original publisher: [Read original article](<https://www.marktechpost.com/2026/10/03/deepseek-harness-v0-2-brings-official-desktop-apps-to-its-open-source-agent-harness/>)

Author: Michal Sutter

Published: 2026-10-04T04:49:42Z

Content type: news

Language: en

Sources: [MarkTechPost](<https://devfeed.tech/sources/marktechpost.md>)

Topics: [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-harness](<https://devfeed.tech/tags/agent-harness.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-shorts](<https://devfeed.tech/tags/ai-shorts.md>), [apple-silicon](<https://devfeed.tech/tags/apple-silicon.md>), [applications](<https://devfeed.tech/tags/applications.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [build](<https://devfeed.tech/tags/build.md>), [context-engineering](<https://devfeed.tech/tags/context-engineering.md>), [desktop](<https://devfeed.tech/tags/desktop.md>), [developer-preview](<https://devfeed.tech/tags/developer-preview.md>), [developer-tools](<https://devfeed.tech/tags/developer-tools.md>), [development](<https://devfeed.tech/tags/development.md>), [editors-pick](<https://devfeed.tech/tags/editors-pick.md>), [for-devs](<https://devfeed.tech/tags/for-devs.md>), [generative-ai](<https://devfeed.tech/tags/generative-ai.md>), [harness](<https://devfeed.tech/tags/harness.md>), [macos](<https://devfeed.tech/tags/macos.md>), [mit](<https://devfeed.tech/tags/mit.md>), [models](<https://devfeed.tech/tags/models.md>), [new-releases](<https://devfeed.tech/tags/new-releases.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [open-source-agent](<https://devfeed.tech/tags/open-source-agent.md>), [staff](<https://devfeed.tech/tags/staff.md>), [tech-news](<https://devfeed.tech/tags/tech-news.md>), [technology](<https://devfeed.tech/tags/technology.md>)

### AI overview

DeepSeek Harness v0.2 is an open-source agent harness with preview desktop apps for macOS and Windows. The release adds plugin management, file and code-change review, scheduled tasks, and support for third-party models through OpenAI-compatible endpoints. The article also describes its plugin-based architecture and an experimental compatibility layer for Claude Code Mods.

### Source excerpt

DeepSeek has released official macOS and Windows desktop apps for DeepSeek Harness v0.2, its MIT-licensed agent harness. The preview adds a plugin manager, file and code-change review, and scheduled Automation Tasks. It also supports non-DeepSeek models through OpenAI-compatible endpoints. The post DeepSeek Harness v0.2 Brings Official Desktop Apps to Its Open-Source Agent Harness appeared first on MarkTechPost.

## How to Start a Gaming Clips Channel for TikTok, Reels, and YouTube Shorts

DevFeed: [How to Start a Gaming Clips Channel for TikTok, Reels, and YouTube Shorts](<https://devfeed.tech/articles/how-to-start-a-gaming-clips-channel-for-tiktok-reels-and-youtube-shorts-84777.md>)

Original publisher: [Read original article](<https://usenaive.ai/blogs/how-to-start-a-gaming-clips-channel>)

Author: Naïve

Published: 2026-10-04T00:00:00Z

Content type: tutorial

Language: en

Sources: [Naïve](<https://devfeed.tech/sources/naive.md>)

Topics: [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [human in the loop agent](<https://devfeed.tech/topics/human-in-the-loop-agent.md>), [browser-game](<https://devfeed.tech/topics/browser-game.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [gaming](<https://devfeed.tech/tags/gaming.md>), [guide](<https://devfeed.tech/tags/guide.md>), [how-to](<https://devfeed.tech/tags/how-to.md>)

### AI overview

This guide explains how to set up a gaming clips channel using naïve's template and AI agent team. Users name streams they own or have permission to clip, set an angle and posting cadence, then review and approve each clip before it is published. The guide also describes spending caps and how a full review queue slows clip production.

### Source excerpt

A guide to running a gaming clips channel: a team of AI agents cuts the clutch plays from streams you own into vertical clips and queues each for your approval.

## Meeting you where you work: Unity Plugin for Grok Build

DevFeed: [Meeting you where you work: Unity Plugin for Grok Build](<https://devfeed.tech/articles/meeting-you-where-you-work-unity-plugin-for-grok-build-79594.md>)

Original publisher: [Read original article](<https://unity.com/blog/unity-plugin-grok>)

Author: Rachel Zhao

Published: 2026-10-01T00:00:00Z

Content type: release

Language: en

Sources: [Unity](<https://devfeed.tech/sources/unity.md>)

Topics: [DeepSeek Harness](<https://devfeed.tech/topics/dsh.md>), [Unity](<https://devfeed.tech/topics/unity.md>)

Tags: [ai-coding-agent](<https://devfeed.tech/tags/ai-coding-agent.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>), [configure](<https://devfeed.tech/tags/configure.md>), [plugin](<https://devfeed.tech/tags/plugin.md>), [terminal](<https://devfeed.tech/tags/terminal.md>), [unity](<https://devfeed.tech/tags/unity.md>)

### AI overview

Unity announced that its official plugin is available for Grok Build, adding 30+ Unity-written skills that were previously available for Claude Code and Codex. The plugins share the same Unity guidance, are maintained from one repository, support Unity 6 and later, and can be installed with one terminal command.

### Source excerpt

The official Unity Plugin now works in Grok Build. Get 30+ Unity-written skills with one terminal command. Unity 6+.

[Next page](<https://devfeed.tech/topics/dsh.md?cursor=WyIyMDI2LTEwLTAxVDAwOjAwOjAwKzAwOjAwIiwgImQ5ZmNiZTVjLWU5YTctNDY4Yi1iZjJiLWFhZWJhOTE4ZGM1OCJd>)