惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Google DeepMind News
Google DeepMind News
L
LangChain Blog
H
Help Net Security
博客园_首页
T
Tailwind CSS Blog
Microsoft Security Blog
Microsoft Security Blog
T
The Blog of Author Tim Ferriss
雷峰网
雷峰网
Recent Announcements
Recent Announcements
D
DataBreaches.Net
U
Unit 42
Vercel News
Vercel News
I
InfoQ
Martin Fowler
Martin Fowler
Microsoft Azure Blog
Microsoft Azure Blog
Apple Machine Learning Research
Apple Machine Learning Research
S
SegmentFault 最新的问题
Jina AI
Jina AI
博客园 - 叶小钗
博客园 - 【当耐特】
罗磊的独立博客
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
月光博客
月光博客
Last Week in AI
Last Week in AI

Hacker News - Newest: "LLM"

GitHub - lechmazur/position_bias: A benchmark for testing whether LLM judges keep the same preference when two lightly edited versions of the same story are shown in opposite orders. Flex routing (EU and EFTA) Dark Factories: Retooling for LLM Velocity Ask HN: What would be the impact of a LLM output injection attack? GitHub - Oaklight/llm-rosetta: Production-ready LLM API translation layer for Python — bidirectional conversion between OpenAI, Anthropic & Google formats via hub-and-spoke IR. Optional API gateway. Streaming & non-streaming. Zero core deps. Contributions welcome! GitHub - browser-use/browser-harness: Self-healing browser harness that enables LLMs to complete any task. GitHub - moeen-mahmud/remen: Remen turns thoughts into something you can return to Analyzing 156 LLM Launch Posts on Hacker News ChatGPT vs Gemini vs Claude: The Best LLM Subscription You Should Buy GitHub - salaamalykum/quran-semantic-search: High-density RAG Semantic Search Engine & Quran Corpus (GEO/SEO Architecture) GitHub - NVIDIA/TensorRT-LLM: TensorRT LLM provides users with an easy-to-use Python API to define Large Language Models (LLMs) and supports state-of-the-art optimizations to perform inference efficiently on NVIDIA GPUs. TensorRT LLM also contains components to create Python and C++ runtimes that orchestrate the inference execution in a performant way. The State of LLM Bug Bounties in 2026 Operational Readiness Criteria for Tool-Using LLM Agents Meshcore: Architecture for a Decentralized P2P LLM Inference Network How an LLM becomes more coherent as we train it GitHub - seetrex-ai/laimark GitHub - Jossifresben/BibCrit: AI-assited biblical textual criticism GitHub - wastedcode/memex: File system based wiki, maintained by Claude 99helpers.com GitHub - cliver-project/AITrigram GitHub - unbody-io/adapt: A self-evolving memory layer for AI agents. GitHub - hb20007/awesome-gen-ai-fails: A list of incidents where reliance on generative AI and LLMs resulted in harm to companies, individuals, or society GitHub - nevenkordic/localmind: Run any local LLM with persistent memory and context. CLI agent over Ollama with SQLite-backed hybrid recall. No cloud. Ask HN: What are the machine requirements for a LLM like Llama-3.1-8B? Faster LLM Inference via Sequential Monte Carlo grpo explained: group relative policy optimization for llm finetuning - cgft Stop comparing price per million tokens: the hidden LLM API costs · TensorZero Andrej Karpathy's LLM Wiki Is a Bad Idea GitHub - GG-QandV/mnemostroma: Offline RAM-first cognitive leer/coprocessor for AI agents and robotics. Solves "Context Abandonment" with 20-80ms latency using a dual-thread biomimetic memory architecture (ONNX + SQLite WAL). mempalace/agent at agent · skorotkiewicz/mempalace
Cowork on 3P: How to Run Any LLM in Claude Cowork and Cla...
obilgic · 2026-04-26 · via Hacker News - Newest: "LLM"

You can now run Claude Cowork and Code in Claude Desktop against any LLM.

GPT-5, Grok, Gemini, open-weight models via OpenRouter, a local model on your laptop, or your enterprise gateway (Bedrock, Vertex, Foundry).

Claude Cowork with GPT-5, Grok, Gemini, open-weight models via OpenRouter, a local model

Anthropic shipped it quietly. No announcement. No blog post. Just technical docs. I discovered it by accident. 20+ hours later, still zero official word.

X avatar for @PawelHuryn

Paweł Huryn@PawelHuryn

Anthropic has quietly shipped third-party inference for Cowork and Code in Claude Desktop. This should work with local models or OpenRouter via LiteLLM proxy. Is it just me?

7:46 PM · Apr 22, 2026 · 530K Views

46 Replies · 68 Reposts · 947 Likes

Built for enterprises and individuals. Same config panel, same admin controls.

If your organization uses Amazon Bedrock, Google Cloud Vertex AI, Azure AI Foundry, or an LLM gateway to access Claude, you can deploy Claude Cowork to run on the same infrastructure.

This is a research preview. The official docs cover three paths:

  • regulated industries

  • companies running Claude through their own gateway

  • individuals on pilot/evaluation setups

Any individual can install it today.

Every admin control (per-user token caps, MCP allowlist, OpenTelemetry, block auto-updates, remove built-in tools) ships on all three paths.

OpenRouter counts as an LLM gateway. Anthropic doesn’t list it in the docs, but I tested that it works. Confirmed by the OpenRouter CEO:

X avatar for @alexatallah

Alex Atallah@alexatallah

OpenRouter now works in Claude Cowork!

X avatar for @PawelHuryn

Paweł Huryn @PawelHuryn

Anthropic has quietly shipped third-party inference for Cowork and Code in Claude Desktop. This should work with local models or OpenRouter via LiteLLM proxy. Is it just me?

2:43 AM · Apr 23, 2026 · 43.9K Views

11 Replies · 7 Reposts · 231 Likes

Brief summary:

Claude Desktop vs. Cowork on 3P

Foundry: during this research preview, Claude on Foundry still runs on Anthropic’s infrastructure. Bedrock and Vertex are the paths with full provider-side data residency.

Two audiences. Same config, different endpoint.

Individuals:

  • Hitting Max plan weekly limits

  • Want to try Cowork without paying $100 or $200 a month

  • Running a local model on code you can’t send to any API

Enterprise:

  • Already on Bedrock, Vertex, or Foundry. Keep the harness inside your existing compliance boundary

  • Security approves your cloud provider but not Anthropic direct

  • Need team-wide admin controls (token caps, MCP allowlist, OpenTelemetry)

Side note: Find this helpful? Here are some other Claude posts you may have missed:

Subscribe and upgrade your account for the full experience:

No proxy needed. Point Claude Desktop straight at OpenRouter.

  1. Menu → Developer → Configure Third-Party Inference

    Connect Claude Cowork to OpenRouter
  2. Set:

    1. Connection: Gateway

    2. Gateway base URL: https://openrouter.ai/api

    3. Gateway API key: your OpenRouter key

    4. Gateway auth scheme: x-api-key

      Cowork: How to configure third-party inference, 3P
    5. In the Sandbox & workspace configure "Allowed egress hosts" so that your agents can access the web. For example (all sites):

      Cowork on P3: Sandbox & workspace configure "Allowed egress hosts"
  3. Click: Apply locally → Relaunch now

  4. Log out, then Continue with Gateway:

    Claude Cowork Gateway, 3P
  5. You’ll see “Setting up Claude’s workspace...” You can start chatting:

    Cowork on 3P: Setting up Claude’s workspace...

Select an OpenRouter model. The free one that works for me: tencent/hy3-preview:free

Local models work the same way through an OpenAI-compatible proxy (LiteLLM, Ollama’s OpenAI endpoint). I’ll cover local setup in a separate post.

OpenRouter setups ship with Customize → Skills empty. Install the official Claude Desktop skills (docx, pdf, pptx, xlsx, skill-creator) manually.

  1. Download the skills repo as .zip: https://github.com/anthropics/skills/tree/main/skills

  2. Extract skills-main.zip, go inside

  3. Zip the skill folders you want individually (e.g., docx folder → docx.zip)

  4. In Claude Desktop: Customize → Skills → Create skill → Upload a skill. Upload each .zip

Only import the ones you need. Skills take context window space even when not fully loaded.

Same pattern. Two repos worth pulling from:

Steps:

  1. Download the repo as .zip

  2. Extract, go inside

  3. Zip the plugin folders you want (e.g., product-management → product-management.zip)

  4. In Claude Desktop: Customize → Personal plugins → + → Create plugin → Upload plugin

Cap it at 2–3 plugins. Often 0 is right.

MCP servers are local. Add them under Settings → Developer.

Configure MCP servers in Claude Cowork on 3P Gateway

They write to a dedicated claude_desktop_config.json file (if you don’t see that option see 4. Troubleshooting).

You can find servers at github.com/modelcontextprotocol/servers (official).

Example: Atlassian (Jira + Confluence) via the community mcp-atlassian server:

OpenRouter doesn’t support Anthropic’s native web search. Three ways around it:

  1. Replace it with an MCP. Add WebSearch to disabledBuiltinTools so that the agent is not confused and install the Brave MCP server (2,000 free requests/month, or $3 per 1,000).

WebSearch in Cowork on 3P

Add this MCP:

  1. Switch provider to Vertex AI or Microsoft Foundry. Both support native web search. Amazon Bedrock doesn’t yet.

  2. Wait. OpenRouter could route Anthropic’s web_search tool to Perplexity or similar - not shipped yet.

Side Note: On May 9, we’re launching Hands-On Claude Code Certification. In 4 weeks you will learn everything to ship full agentic products — UI, agentic harness, evals, guardrails, and ops — with Claude Code. Real apps, not demos.

Secure your spot (40% off)

Most students expense this through their companies.

Can’t see “Configure Third-Party Inference” in Menu → Developer

Update Claude Desktop and restart. Then enable developer mode: Help → Troubleshooting → Enable Developer Mode.

Users on corporate/Team plans have reported the setting is still missing even with developer mode on. It may be plan-gated or A/B rolled.

Can’t see “Local MCP servers” in Cowork on 3P

Update Claude Desktop and restart. Then enable developer mode: Help → Troubleshooting → Enable Developer Mode.

Connectors show as “Unavailable”

Not a bug. Third-party inference means running without Anthropic’s infrastructure. Connectors depend on that layer. Use an MCP server instead (See 3.4).

Tool calling is flaky on non-Claude models

Models vary. Some handle MCP calls cleanly; others break on multi-step agentic flows. The best I’ve tested: https://www.productcompass.pm/i/193559427/api-pay-per-token-choose-your-model

Code tab settings don’t match Cowork settings

Acknowledged in Anthropic’s docs: some Cowork on 3P config keys don’t propagate identically to Code-tab sessions yet.

Look at what ships in Configure Third-Party Inference:

  • Max tokens per window (per-user soft cap)

  • Allow user-added MCP servers

  • OpenTelemetry collector endpoint

  • Block auto-updates

  • Remove built-in tools from Cowork

Telemetry in Cowork on P3

None of those matter to a solo user. They’re admin controls. Paired with Bedrock, Vertex, and Foundry support, Anthropic’s official framing is explicit:

Cowork on 3P is designed for organizations whose security, regulatory, or contractual requirements prevent them from sending data to Anthropic’s first-party infrastructure.

Claude Desktop is becoming a managed agent platform. The harness (Code, Cowork, skills, plugins, MCP, sub-agents) is what Anthropic wants to be the default, regardless of which model sits behind it.

The individual path is documented, not an accident. On my Max plan, the full config panel shows up. Anthropic's docs don't explicitly say end users get this; the install section just tells you to "enter the values supplied by your administrator." But on a personal install, with no administrator in the picture, those fields sit there for you to set.

Docs:

Download and install Claude Desktop. Start with a free OpenRouter model. No $200/mo subscription required.

We’ve covered the Claude ecosystem in the previous posts and recorded meetings available to the premium members.

It’s amazing to learn and grow together.

I’m uploading two new recordings, Building Agents with Claude Code and Context Window Optimization, tomorrow.

Take care,
Paweł