[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-71961":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":25,"hasPages":23,"topics":26,"createdAt":10,"pushedAt":10,"updatedAt":35,"readmeContent":36,"aiSummary":37,"trendingCount":16,"starSnapshotCount":16,"syncStatus":38,"lastSyncTime":39,"discoverSource":40},71961,"open-swe","langchain-ai\u002Fopen-swe","langchain-ai","An Open-Source Asynchronous Coding Agent","",null,"Python",9944,1130,52,6,0,20,51,164,60,40.16,"MIT License",false,"main",true,[27,28,29,30,31,32,33,34],"agent","agents","ai","anthropic","claudecode","llm","llms","openai","2026-06-12 02:02:56","\u003Cdiv align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fopen-swe\">\n    \u003Cpicture>\n      \u003Csource media=\"(prefers-color-scheme: dark)\" srcset=\"static\u002Fdark.svg\">\n      \u003Csource media=\"(prefers-color-scheme: light)\" srcset=\"static\u002Flight.svg\">\n      \u003Cimg alt=\"Open SWE Logo\" src=\"static\u002Fdark.svg\" width=\"35%\">\n    \u003C\u002Fpicture>\n  \u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003Cdiv align=\"center\">\n  \u003Ch3>Open-source framework for building your org's internal coding agent.\u003C\u002Fh3>\n\u003C\u002Fdiv>\n\n\u003Cdiv align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fopensource.org\u002Flicenses\u002FMIT\" target=\"_blank\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Flicense\u002Flangchain-ai\u002Fopen-swe\" alt=\"License\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fopen-swe\u002Fstargazers\" target=\"_blank\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Flangchain-ai\u002Fopen-swe\" alt=\"GitHub Stars\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Flanggraph\" target=\"_blank\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FBuilt%20on-LangGraph-blue\" alt=\"Built on LangGraph\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\" target=\"_blank\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FBuilt%20on-Deep%20Agents-blue\" alt=\"Built on Deep Agents\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fx.com\u002Flangchain\" target=\"_blank\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Ftwitter\u002Furl\u002Fhttps\u002Ftwitter.com\u002Flangchain.svg?style=social&label=Follow%20%40LangChain\" alt=\"Twitter \u002F X\">\u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003Cbr>\n\nElite engineering orgs like Stripe, Ramp, and Coinbase are building their own internal coding agents — Slackbots, CLIs, and web apps that meet engineers where they already work. These agents are connected to internal systems with the right context, permissioning, and safety boundaries to operate with minimal human oversight.\n\nOpen SWE is the open-source version of this pattern. Built on [LangGraph](https:\u002F\u002Flangchain-ai.github.io\u002Flanggraph\u002F) and [Deep Agents](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents), it gives you the same architecture those companies built internally: cloud sandboxes, Slack and Linear invocation, subagent orchestration, and automatic PR creation — ready to customize for your own codebase and workflows.\n\n> [!NOTE]\n> 💬 Read the **announcement blog post [here](https:\u002F\u002Fblog.langchain.com\u002Fopen-swe-an-open-source-framework-for-internal-coding-agents\u002F)**\n\n---\n\n## Architecture\n\nOpen SWE makes the same core architectural decisions as the best internal coding agents. Here's how it maps to the patterns described in [this overview](https:\u002F\u002Fx.com\u002Fkishan_dahya\u002Fstatus\u002F2028971339974099317) of Stripe's Minions, Ramp's Inspect, and Coinbase's Cloudbot:\n\n### 1. Agent Harness — Composed on Deep Agents\n\nRather than forking an existing agent or building from scratch, Open SWE **composes** on the [Deep Agents](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents) framework — similar to how Ramp built on top of OpenCode. This gives you an upgrade path (pull in upstream improvements) while letting you customize the orchestration, tools, and middleware for your org.\n\n```python\ncreate_deep_agent(\n    model=\"openai:gpt-5.5\",\n    system_prompt=construct_system_prompt(...),\n    tools=[http_request, fetch_url, linear_comment, slack_thread_reply],\n    backend=sandbox_backend,\n    middleware=[ToolErrorMiddleware(), check_message_queue_before_model, ...],\n)\n```\n\n### 2. Sandbox — Isolated Cloud Environments\n\nEvery task runs in its own **isolated cloud sandbox** — a remote Linux environment with full shell access. The repo is cloned in, the agent gets full permissions, and the blast radius of any mistake is fully contained. No production access, no confirmation prompts.\n\nOpen SWE supports multiple sandbox providers out of the box — [Modal](https:\u002F\u002Fmodal.com\u002F), [Daytona](https:\u002F\u002Fwww.daytona.io\u002F), [Runloop](https:\u002F\u002Fwww.runloop.ai\u002F), and [LangSmith](https:\u002F\u002Fsmith.langchain.com\u002F) — and you can plug in your own. See the [Customization Guide](CUSTOMIZATION.md#1-sandbox) for details.\n\nThis follows the principle all three companies converge on: **isolate first, then give full permissions inside the boundary.**\n\n- Each thread gets a persistent sandbox (reused across follow-up messages)\n- Sandboxes auto-recreate if they become unreachable\n- Multiple tasks run in parallel — each in its own sandbox, no queuing\n\n### 3. Tools — Curated, Not Accumulated\n\nStripe's key insight: *tool curation matters more than tool quantity.* Open SWE follows this principle with a small, focused toolset:\n\n| Tool | Purpose |\n|---|---|\n| `execute` | Shell commands in the sandbox |\n| `fetch_url` | Fetch web pages as markdown |\n| `http_request` | API calls (GET, POST, etc.) |\n| `linear_comment` | Post updates to Linear tickets |\n| `slack_thread_reply` | Reply in Slack threads |\n\nGitHub operations are performed with `GH_TOKEN=dummy gh` inside the sandbox, backed by the LangSmith proxy. Plus the built-in Deep Agents tools: `read_file`, `write_file`, `edit_file`, `ls`, `glob`, `grep`, `write_todos`, and `task` (subagent spawning).\n\n### 4. Context Engineering — AGENTS.md + Source Context\n\nOpen SWE gathers context from two sources:\n\n- **`AGENTS.md`** — If the repo contains an `AGENTS.md` file at the root, it's read from the sandbox and injected into the system prompt. This is your repo-level equivalent of Stripe's rule files: encoding conventions, testing requirements, and architectural decisions that every agent run should follow.\n- **Source context** — The full Linear issue (title, description, comments) or Slack thread history is assembled and passed to the agent, so it starts with rich context rather than discovering everything through tool calls.\n\n### 5. Orchestration — Subagents + Middleware\n\nOpen SWE's orchestration has two layers:\n\n**Subagents:** The Deep Agents framework natively supports spawning child agents via the `task` tool. The main agent can fan out independent subtasks to isolated subagents — each with its own middleware stack, todo list, and file operations. This is similar to Ramp's child sessions for parallel work.\n\n**Middleware:** Deterministic middleware hooks run around the agent loop:\n\n- **`check_message_queue_before_model`** — Injects follow-up messages (Linear comments or Slack messages that arrive mid-run) before the next model call. You can message the agent while it's working and it'll pick up your input at its next step.\n- **`notify_step_limit_reached`** — After-agent hook that posts a Slack reply when the agent hits the model-call limit, so users get a clear signal instead of silence.\n- **`ToolErrorMiddleware`** — Catches and handles tool errors gracefully.\n\n### 6. Invocation — Slack, Linear, and GitHub\n\nAll three companies in the article converge on **Slack as the primary invocation surface**. Open SWE does the same:\n\n- **Slack** — Mention the bot in any thread. Supports `repo:owner\u002Fname` syntax to specify which repo to work on. The agent replies in-thread with status updates and PR links.\n- **Linear** — Comment `@openswe` on any issue. The agent reads the full issue context, reacts with 👀 to acknowledge, and posts results back as comments.\n- **GitHub** — Tag `@openswe` in PR comments on agent-created PRs to have it address review feedback and push fixes to the same branch.\n\nEach invocation creates a deterministic thread ID, so follow-up messages on the same issue or thread route to the same running agent.\n\n### 7. Validation — Prompt-Driven\n\nThe agent is instructed to run linters, formatters, and tests before committing, and is responsible end-to-end for committing, pushing, opening\u002Fupdating the draft PR, and replying in the source channel.\nThis is an area where you can extend Open SWE for your org: add deterministic CI checks, visual verification, or review gates as additional middleware. See the [Customization Guide](CUSTOMIZATION.md#6-middleware) for how.\n\n---\n\n## Comparison\n\n| Decision | Open SWE | Stripe (Minions) | Ramp (Inspect) | Coinbase (Cloudbot) |\n|---|---|---|---|---|\n| **Harness** | Composed (Deep Agents\u002FLangGraph) | Forked (Goose) | Composed (OpenCode) | Built from scratch |\n| **Sandbox** | Pluggable (Modal, Daytona, Runloop, etc.) | AWS EC2 devboxes (pre-warmed) | Modal containers (pre-warmed) | In-house |\n| **Tools** | ~15, curated | ~500, curated per-agent | OpenCode SDK + extensions | MCPs + custom Skills |\n| **Context** | AGENTS.md + issue\u002Fthread | Rule files + pre-hydration | OpenCode built-in | Linear-first + MCPs |\n| **Orchestration** | Subagents + middleware | Blueprints (deterministic + agentic) | Sessions + child sessions | Three modes |\n| **Invocation** | Slack, Linear, GitHub | Slack + embedded buttons | Slack + web + Chrome extension | Slack-native |\n| **Validation** | Prompt-driven | 3-layer (local + CI + 1 retry) | Visual DOM verification | Agent councils + auto-merge |\n\n---\n\n## Features\n\n- **Trigger from Linear, Slack, or GitHub** — mention `@openswe` in a comment to kick off a task\n- **Instant acknowledgement** — reacts with 👀 the moment it picks up your message\n- **Message it while it's running** — send follow-up messages mid-task and it'll pick them up before its next step\n- **Run multiple tasks in parallel** — each task runs in its own isolated cloud sandbox\n- **GitHub OAuth built-in** — authenticates with your GitHub account automatically\n- **Opens PRs automatically** — commits changes and opens a draft PR when done, linked back to your ticket\n- **Subagent support** — the agent can spawn child agents for parallel subtasks\n\n---\n\n## Getting Started\n\n- **[Installation Guide](INSTALLATION.md)** — GitHub App creation, LangSmith, Linear\u002FSlack\u002FGitHub triggers, and production deployment\n- **[Customization Guide](CUSTOMIZATION.md)** — swap the sandbox, model, tools, triggers, system prompt, and middleware for your org\n\n## License\n\nMIT\n","Open SWE 是一个开源的异步编码代理框架，用于构建组织内部的编码助手。该项目基于 Python 开发，利用了 LangGraph 和 Deep Agents 技术栈，支持云沙箱、Slack 和 Linear 调用、子代理协调以及自动 PR 创建等功能，为开发者提供了一个可定制的基础架构来集成到现有工作流中。适用于需要提升开发效率、自动化代码生成与审查的企业级应用场景，特别是那些希望在已有系统基础上快速搭建智能辅助工具的技术团队。",2,"2026-06-11 03:39:41","high_star"]