[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-92336":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":15,"subscribersCount":15,"size":15,"stars1d":15,"stars7d":15,"stars30d":16,"stars90d":15,"forks30d":15,"starsTrendScore":15,"compositeScore":17,"rankGlobal":10,"rankLanguage":10,"license":18,"archived":19,"fork":19,"defaultBranch":20,"hasWiki":19,"hasPages":21,"topics":22,"createdAt":10,"pushedAt":10,"updatedAt":23,"readmeContent":24,"aiSummary":25,"trendingCount":15,"starSnapshotCount":15,"syncStatus":26,"lastSyncTime":27,"discoverSource":28},92336,"OpenClaudeScience","qzzqzzb\u002FOpenClaudeScience","qzzqzzb","Bringing the Claude Science experience to the open-source world.","https:\u002F\u002Finternagents.github.io\u002F",null,"TypeScript",69,9,52,0,14,44.4,"MIT License",false,"main",true,[],"2026-07-22 04:02:05","\u003Cdiv align=\"center\">\n  \u003Cp>\n    \u003Cimg src=\".\u002Fdocs\u002Fassets\u002Freadme\u002Finternagents-banner.png\" alt=\"InternAgentS banner\" width=\"100%\">\n  \u003C\u002Fp>\n\n  \u003Ch1 align=\"center\">InternAgentS: Bringing the Claude Science experience to the open-source world.\u003C\u002Fh1>\n  \u003Cp align=\"center\">\n    Built on DeepAgents and LangGraph to extend research agent runtimes across project context, files, skills, remote resources, and human approvals.\n  \u003C\u002Fp>\n  \u003Cp align=\"center\">\n    \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fqzzqzzb\u002FOpenClaudeScience\u002Fstargazers\">\u003Cimg alt=\"GitHub stars\" src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fqzzqzzb\u002FOpenClaudeScience?style=social\">\u003C\u002Fa>\n    \u003Cimg alt=\"Python\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPython-3.11%2B-3776AB?logo=python&logoColor=white\">\n    \u003Cimg alt=\"Next.js\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNext.js-16-black?logo=nextdotjs\">\n    \u003Cimg alt=\"LangGraph\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLangGraph-runtime-1f6feb\">\n    \u003Cimg alt=\"DeepAgents\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDeepAgents-agent%20runtime-4f46e5\">\n    \u003Cimg alt=\"License\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-green\">\n    \u003Cimg alt=\"Status\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fstatus-active%20development-0f766e\">\n  \u003C\u002Fp>\n  \u003Cp align=\"center\">\n    \u003Cstrong>English\u003C\u002Fstrong> | \u003Ca href=\".\u002FREADME_CN.md\">简体中文\u003C\u002Fa>\n  \u003C\u002Fp>\n  \u003Cp>\n    \u003Ca href=\"#highlights\">Highlights\u003C\u002Fa>\n    · \u003Ca href=\"#example-workflows\">Workflows\u003C\u002Fa>\n    · \u003Ca href=\"#feature-overview\">Feature Overview\u003C\u002Fa>\n    · \u003Ca href=\"#quick-start\">Quick Start\u003C\u002Fa>\n    · \u003Ca href=\"#security-and-privacy\">Security\u003C\u002Fa>\n    · \u003Ca href=\"#architecture\">Architecture\u003C\u002Fa>\n    · \u003Ca href=\"#development\">Development\u003C\u002Fa>\n    · \u003Ca href=\"#license\">License\u003C\u002Fa>\n  \u003C\u002Fp>\n\u003C\u002Fdiv>\n\nInternAgentS is a local-first research workspace built on DeepAgents\u002FLangGraph,\nwith project runtimes, skills, remote resources, and approval flows designed for\nscientific work.\n\n## Highlights\n\n- **DeepAgents\u002FLangGraph, extended for research**: InternAgentS adapts the\n  runtime, workspace, skills, tools, and approval flow around scientific\n  projects.\n- **Remote environments without the ceremony**: connect SSH workspaces, sync\n  remote runtimes, inspect logs, and approve remote compute jobs from the\n  conversation.\n- **Science skills out of the box**: literature search, result analysis,\n  figures, paper writing, documents, slides, and domain workflows are reusable\n  skills.\n- **MCP\u002FSCP, plus your choice of model**: connect external tools through\n  MCP\u002FSCP, and use cloud models, private gateways, or local model servers.\n- **Local-first data control**: project files, secrets, and runtime state stay\n  on machines you control by default, without requiring Claude, Claude Science,\n  or any fixed cloud service.\n- **Built for real scientific files**: browse, search, preview, and reference\n  PDFs, Office files, images, molecular structures, scientific data outputs, and\n  generated artifacts.\n\n## Example Workflows\n\n- Paper and report triage: attach papers or Markdown reports, ask the agent to\n  summarize claims, extract assumptions, compare methods, and leave generated\n  notes in the project.\n- Scientific artifact inspection: browse project files, preview PDFs, Office\n  files, images, molecular structures, and scientific data outputs, then ask the\n  agent to explain what changed.\n- Experiment and code iteration: ask the agent to inspect code, run local\n  commands, create result files, and summarize outputs with links back to the\n  files in the workspace.\n- Skill-guided sessions: enable reusable skills for literature search, result\n  analysis, figures, documents, slides, or domain-specific research workflows.\n- Remote compute handoff: register a Linux SSH host, review the proposed compute\n  job in chat, approve it, and let the local backend harvest configured outputs.\n\n### Analyzing Potential Lattice Distortions in PbTiO3 Perovskites Under A-Site\u002FB-Site Substitution\n\n![PbTiO3 A-site\u002FB-site substitution lattice distortion analysis](.\u002Fdocs\u002Fassets\u002Freadme\u002Fpbtio3-ab-site-substitution.gif)\n\n### Caffeine Computational Chemistry Study\n\n![Caffeine computational chemistry study](.\u002Fdocs\u002Fassets\u002Freadme\u002Fcaffeine-computational-chemistry.gif)\n\n### Construct simplified 2D\u002F3D models of a Y-shaped microfluidic mixer\n\n![Construct simplified 2D\u002F3D models of a Y-shaped microfluidic mixer](.\u002Fdocs\u002Fassets\u002Freadme\u002Fy-shaped-microfluidic-mixer.gif)\n\n## Feature Overview\n\nInternAgentS brings chat, project sessions, file browsing, and local runtime\nstatus into a single research workbench. The right panel keeps project files and\nartifacts visible while the center conversation stays focused on the current\ntask.\n\n![InternAgentS workspace preview](.\u002Fdocs\u002Fassets\u002Freadme\u002Fworkspace-preview-en.jpeg)\n\n### Local-First Research Workspace\n\nInternAgentS is organized as a three-panel workspace:\n\n| Area | What it does |\n| --- | --- |\n| Left sidebar | project navigation, sessions, settings, and skill entry points |\n| Center | chat, composer, attachments, mentions, and agent progress |\n| Right panel | project files, previews, provenance, runtime info, and connector context |\n\nProject files are accessed through the workspace API rather than direct UI file\nsystem calls. The file panel supports directory navigation, grid\u002Flist views,\nsearch, and previews for common research artifacts.\n\n### Skills and Science Capability Library\n\nSkills are reusable capabilities that can be enabled for an agent or session.\nInternAgentS searches shared user catalogs first, then project catalogs:\n\n```text\n~\u002F.internagents\u002Fmyskills\n~\u002F.internagents\u002Fimported-skills\nskills\n.internagents\u002Fimported-skills\n```\n\nThe settings UI supports built-in skills, imported skills, and science skills.\nImported skills are copied into a user-level catalog so the same capability can\nbe reused across multiple projects.\n\n### Model, Authorization, and Appearance Settings\n\nThe unified settings page manages:\n\n- model provider, Base URL, API key, and model ID\n- project directory\n- Linux SSH compute host registration and job activity\n- tool-call authorization mode\n- language and appearance\n- archived conversations\n- skills and connector configuration\n\nThe UI includes both Chinese and English copy.\n\n### MCP and SCP Connectors\n\nInternAgentS can load external tools through MCP server configuration and can\nprepare SCP Hub access for science skill workflows.\n\nLocal MCP config locations:\n\n```text\n~\u002F.deepagents\u002F.mcp.json\n\u003Crepo>\u002F.deepagents\u002F.mcp.json\n\u003Crepo>\u002F.mcp.json\nINTERNAGENT_MCP_CONFIG_FILE\n```\n\nConnector secrets, private commands, headers, and endpoints should stay local.\n\n### Linux SSH Compute Jobs\n\nInternAgentS has an experimental Linux-only SSH compute provider. This is\nseparate from SSH remote runtime setup: the local backend keeps the current\nconversation session and submits detached jobs to a registered Linux SSH host.\n\nCurrent scope:\n\n- Linux hosts only.\n- SSH hosts are registered by `Host` alias from the local `~\u002F.ssh\u002Fconfig`.\n  Address, user, port, `ProxyJump`, and key settings come from OpenSSH.\n- Jobs run as detached `bash` processes under a per-job scratch directory.\n- Job status is polled over SSH; outputs matching configured globs are harvested\n  back as base64 payloads when they fit under the configured size cap.\n- Settings > Compute registers and probes SSH hosts. Job submission happens from\n  the conversation when the agent proposes a remote compute tool call.\n- Proposed remote compute calls appear as permission cards in chat. The user\n  must approve the card before the local backend submits the SSH job.\n\nLocal compute state lives under `.internagents\u002Fcompute\u002F`, which is ignored by\ngit. The local API surface is:\n\n```text\nGET  \u002Fapi\u002Fcompute\u002Fssh-hosts\nPOST \u002Fapi\u002Fcompute\u002Fssh-hosts\nGET  \u002Fapi\u002Fcompute\u002Fremote-jobs\nPOST \u002Fapi\u002Fcompute\u002Fremote-jobs\nGET  \u002Fapi\u002Fcompute\u002Fremote-jobs\u002F:jobId\n```\n\nAPI calls require the local token stored at `.internagents\u002Fcompute\u002Fapi-token`:\n\n```bash\nTOKEN=\"$(cat .internagents\u002Fcompute\u002Fapi-token)\"\ncurl -X POST http:\u002F\u002F127.0.0.1:3000\u002Fapi\u002Fcompute\u002Fssh-hosts \\\n  -H 'Content-Type: application\u002Fjson' \\\n  -H \"X-InternAgentS-Compute-Token: $TOKEN\" \\\n  -d '{\"host\":\"my-linux-host\",\"notes\":\"Use sbatch on gpu partition; conda envs live under ~\u002Fenvs.\"}'\n```\n\n## Quick Start\n\n### Requirements\n\n- Python 3.11+\n- Node.js and npm. The UI uses `ui\u002Fpackage-lock.json` as the canonical lockfile.\n- An OpenAI-compatible model endpoint, or the option to configure one later\n\n### Start the Workbench\n\n```bash\ncp .env.example .env\n.\u002Fscripts\u002Fdev.sh\n```\n\nThe launcher prepares the local environment and starts three services:\n\nOn first run, it creates `.venv`, installs the Python package in editable mode,\nand runs `npm install --legacy-peer-deps --ignore-scripts` in `ui\u002F`. Use\n`INTERNAGENTS_SKIP_INSTALL=1` only after these dependencies are already present.\n\n| Service | Default URL | Purpose |\n| --- | --- | --- |\n| UI | `http:\u002F\u002F127.0.0.1:3000` | Next.js workbench |\n| Coordinator | `http:\u002F\u002F127.0.0.1:2024` | LangGraph API for the workbench frontend |\n| Local runtime | `http:\u002F\u002F127.0.0.1:22024` | Project-scoped DeepAgent runtime |\n\nOpen:\n\n```text\nhttp:\u002F\u002F127.0.0.1:3000\u002F?assistantId=agent_local\n```\n\nLogs are written to:\n\n```text\n.internagents\u002Flogs\u002Fbackend.log\n.internagents\u002Flogs\u002Flocal-runtime.log\n.internagents\u002Flogs\u002Fui.log\n```\n\nPress `Ctrl+C` in the launcher terminal to stop the services started by the\nscript.\n\n### Configure a Model\n\nYou can configure a model during first setup, or skip it and return later from\nSettings. For an OpenAI-compatible endpoint:\n\n```env\nINTERNAGENTS_MODEL_PROVIDER=openai_compatible\nOPENAI_BASE_URL=https:\u002F\u002Fapi.example.com\u002Fv1\nOPENAI_API_KEY=sk-...\nDEEPAGENT_MODEL=your-model-id\n```\n\nDeepSeek's official OpenAI-compatible endpoint can also be configured with\nprovider-specific aliases:\n\n```env\nDEEPSEEK_API_KEY=\nDEEPSEEK_BASE_URL=https:\u002F\u002Fapi.deepseek.com\nDEEPSEEK_MODEL=deepseek-chat\n```\n\nWhen the OpenAI-compatible provider is selected, `DEEPSEEK_API_KEY`,\n`DEEPSEEK_BASE_URL`, and `DEEPSEEK_MODEL` are treated as aliases for the\ncorresponding OpenAI-compatible API key, base URL, and model.\n\nKeep API keys and machine-specific paths in local `.env` or runtime config\nfiles. Do not commit secrets.\n\n### Useful Overrides\n\n```bash\nINTERNAGENTS_UI_PORT=3001 .\u002Fscripts\u002Fdev.sh\nINTERNAGENTS_BACKEND_PORT=2025 .\u002Fscripts\u002Fdev.sh\nINTERNAGENTS_OPEN_BROWSER=0 .\u002Fscripts\u002Fdev.sh\nINTERNAGENTS_SKIP_INSTALL=1 .\u002Fscripts\u002Fdev.sh\n```\n\n## Security and Privacy\n\nInternAgentS is local-first by default. Project files are accessed through the\nworkspace API, and runtime state is kept under local directories such as\n`.internagents\u002F`.\n\n- Keep model API keys, MCP headers, SCP Hub keys, server addresses, SSH aliases,\n  and machine-specific paths in local `.env` or runtime config files.\n- Do not commit `.env`, `internagent.resources.local.json`, private SSH\n  material, logs, pids, uploads, LangGraph state, or active skill runtime\n  directories.\n- Tool-call authorization modes can require approval before file writes or other\n  actions. SSH compute jobs always appear as approval cards before submission.\n- When connecting to a remote Agent service, review that service endpoint first:\n  the remote service owns its own workspace, tools, and resource policy.\n- Connector configuration should keep secrets local. Shared examples should be\n  sanitized and should prefer placeholder endpoints and keys.\n\n## Architecture\n\n```mermaid\nflowchart LR\n  Browser[\"Browser UI\u003Cbr\u002F>Next.js\"] --> Coordinator[\"LangGraph coordinator\u003Cbr\u002F>agent.py\"]\n  Coordinator --> Runtime[\"Local runtime\u003Cbr\u002F>DeepAgent\"]\n  Runtime --> Workspace[\"Project workspace\u003Cbr\u002F>files, shell, skills\"]\n  Runtime --> MCP[\"MCP servers\"]\n  Runtime --> SCP[\"SCP Hub\"]\n```\n\n## Repository Map\n\n```text\nagent.py                         LangGraph entrypoint shim\ndeepagent.config.json            local backend, skills, model, and UI defaults\ninternagents\u002F                    graph assembly, backend adapters, middleware, tools, and resource loading\nscripts\u002Fdev.sh                   one-command local development launcher\nui\u002F                              Next.js workbench UI\nskills\u002F                          bundled project skills\ndocs\u002F                            user guides and design notes\n```\n\n## Development\n\nRun these checks before opening a pull request:\n\n```bash\ngit diff --check\npython3 -m json.tool deepagent.config.json >\u002Fdev\u002Fnull\npython3 -m json.tool internagent.resources.json >\u002Fdev\u002Fnull\npython3 -m json.tool ui\u002Fdeepagent-ui.config.json >\u002Fdev\u002Fnull\nnpm --prefix ui run lint\n(cd ui && npx tsc --noEmit)\nnpm --prefix ui run build\n```\n\nFor Python backend changes:\n\n```bash\n.venv\u002Fbin\u002Fpython -m compileall agent.py internagents\n.venv\u002Fbin\u002Fpython -c \"import agent; print(agent.MODEL)\"\n```\n\n## Contributing\n\nInternAgentS is shaped as an open research tool. Helpful contributions include:\n\n- bug reports with clear reproduction steps\n- UI polish that keeps existing workflows stable\n- new skills with examples and safe defaults\n- connector integrations that keep secrets local\n- documentation for installation, configuration, and research workflows\n\nPlease keep changes scoped. DeepAgents is treated as an external SDK, so\nInternAgentS should extend it through public APIs, adapters, middleware, tools,\nand local resource configuration rather than patching SDK internals.\n\n## License\n\nInternAgentS is released under the [MIT License](LICENSE).\n\n## Roadmap Notes\n\nNear-term areas of work:\n\n- clearer skill marketplace and installation flow\n- stronger MCP and SCP configuration UX\n- richer previews for scientific artifacts\n- better remote resource management\n- packaged desktop workflows\n\n## Contributors\n\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fqzzqzzb\u002FOpenClaudeScience\u002Fgraphs\u002Fcontributors\">\n  \u003Cimg src=\"https:\u002F\u002Fcontrib.rocks\u002Fimage?repo=qzzqzzb\u002FOpenClaudeScience\" alt=\"Contributors\" \u002F>\n\u003C\u002Fa>\n","InternAgentS 是一个面向科研场景的开源研究智能体工作台，基于 DeepAgents 和 LangGraph 构建，支持项目上下文管理、文件处理、技能调用、远程计算环境接入及人工审批流程。核心特点包括本地优先的数据控制、开箱即用的科研技能（如文献检索、图表生成、论文撰写）、对 PDF\u002FOffice\u002F科学数据等多格式文件的原生支持，以及通过 MCP\u002FSCP 协议集成外部工具的能力。适用于高校实验室、独立研究员及科研团队开展本地化、可审计、隐私敏感的自动化科研协作任务。",2,"2026-07-08 04:30:09","CREATED_QUERY"]