[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-80891":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":8,"htmlUrl":8,"language":9,"languages":8,"totalLinesOfCode":8,"stars":10,"forks":11,"watchers":12,"openIssues":13,"contributorsCount":13,"subscribersCount":13,"size":13,"stars1d":13,"stars7d":13,"stars30d":14,"stars90d":13,"forks30d":13,"starsTrendScore":13,"compositeScore":15,"rankGlobal":8,"rankLanguage":8,"license":16,"archived":17,"fork":17,"defaultBranch":18,"hasWiki":17,"hasPages":17,"topics":19,"createdAt":8,"pushedAt":8,"updatedAt":20,"readmeContent":21,"aiSummary":22,"trendingCount":13,"starSnapshotCount":13,"syncStatus":11,"lastSyncTime":23,"discoverSource":24},80891,"Intern-Atlas","OpenRaiser\u002FIntern-Atlas","OpenRaiser",null,"Python",35,2,34,0,1,38.53,"MIT License",false,"main",[],"2026-06-12 04:01:30","# Intern Atlas\n\nIntern Atlas is a small toolkit for building a local methodology-evolution graph\nfrom a folder of papers. Give it PDFs or title\u002Fabstract files, configure an\nOpenAI-compatible chat API, and it writes a SQLite graph that can be queried by\na local FastAPI service.\n\nThis repository is intentionally not the Intern Atlas website. It contains the\nreusable local builder, local API, and a small client for the hosted Intern Atlas\nAPI.\n\n## For Online API\n\nThe API document is here: https:\u002F\u002Fjcnc4nyy2d5p.feishu.cn\u002Fwiki\u002FEL1sw9XqAizaCYkGoxpcPAoNnKc\n\n## What It Builds\n\nThe local graph has:\n\n- `papers`: paper nodes from PDFs or metadata files.\n- `methods`: normalized method names extracted from each paper.\n- `paper_methods`: which papers introduce, use, extend, or compare methods.\n- `citations`: method-evolution edges such as `extends`, `improves`,\n  `replaces`, `adapts`, `combines`, and `uses_component`.\n\nThe output is a SQLite database and, optionally, a JSON export.\n\n## Install\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FOpenRaiser\u002FIntern-Atlas.git\ncd Intern-Atlas\npython -m venv .venv\nsource .venv\u002Fbin\u002Factivate\npip install -e .\n```\n\n## Configure An LLM\n\nThe builder uses an OpenAI-compatible `\u002Fchat\u002Fcompletions` endpoint.\n\n```bash\nexport S4S_LLM_BASE_URL=\"https:\u002F\u002Fyour-openai-compatible-host\u002Fv1\"\nexport S4S_LLM_API_KEY=\"YOUR_API_KEY\"\nexport S4S_LLM_MODELS=\"gpt-4o-mini\"\n```\n\nMultiple fallback models are comma-separated:\n\n```bash\nexport S4S_LLM_MODELS=\"model-a,model-b,model-c\"\n```\n\nFor a smoke test without any network call, add `--no-llm`. The heuristic mode\nis useful for checking input format, but LLM mode gives much better bottlenecks\nand mechanisms.\n\n## Build A Graph\n\nTXT input:\n\n```bash\nintern-atlas build \\\n  --input examples\u002Fpapers.txt \\\n  --out outputs\u002Flocal_graph.db \\\n  --json outputs\u002Flocal_graph.json\n```\n\nPDF input:\n\n```bash\nintern-atlas build \\\n  --pdf-dir .\u002Fpapers \\\n  --out outputs\u002Flocal_graph.db \\\n  --json outputs\u002Flocal_graph.json \\\n  --max-pdf-pages 8\n```\n\nSmoke test without LLM:\n\n```bash\nintern-atlas build \\\n  --input examples\u002Fpapers.txt \\\n  --out outputs\u002Fsmoke.db \\\n  --json outputs\u002Fsmoke.json \\\n  --no-llm\n```\n\n## Serve The Local API\n\n```bash\nintern-atlas serve --db outputs\u002Flocal_graph.db --host 127.0.0.1 --port 8000\n```\n\nThen open the local graph workspace:\n\n- `http:\u002F\u002F127.0.0.1:8000\u002F`\n\nAPI docs and health checks:\n\n- `http:\u002F\u002F127.0.0.1:8000\u002Fapi\u002Fdocs`\n- `http:\u002F\u002F127.0.0.1:8000\u002Fapi\u002Fhealth`\n- `http:\u002F\u002F127.0.0.1:8000\u002Fapi\u002Fstats`\n\nExample:\n\n```bash\ncurl \"http:\u002F\u002F127.0.0.1:8000\u002Fapi\u002Fpapers\u002Fsearch?q=attention\"\ncurl \"http:\u002F\u002F127.0.0.1:8000\u002Fapi\u002Fmethods?q=Transformer\"\ncurl -X POST \"http:\u002F\u002F127.0.0.1:8000\u002Fapi\u002Fv1\u002Fevidence\u002Fcontext\" \\\n  -H \"Content-Type: application\u002Fjson\" \\\n  -d '{\"query\":\"efficient attention\",\"mode\":\"deep\",\"year_from\":2020,\"max_papers\":40,\"max_edges\":80}'\ncurl -X POST \"http:\u002F\u002F127.0.0.1:8000\u002Fapi\u002Fassist\u002Fcontext\" \\\n  -H \"Content-Type: application\u002Fjson\" \\\n  -d '{\"query\":\"efficient attention\",\"max_nodes\":80}'\n```\n\nFor LLM and agent integrations, prefer the versioned API:\n\n- `POST \u002Fapi\u002Fv1\u002Fevidence\u002Fcontext`\n- `GET \u002Fapi\u002Fv1\u002Fmethods\u002Fsearch`\n- `GET \u002Fapi\u002Fv1\u002Fevolution\u002Fedges`\n- `GET \u002Fapi\u002Fv1\u002Fpapers\u002F{paper_id}\u002Fneighborhood`\n- `GET \u002Fapi\u002Fv1\u002Fllm\u002Ftools`\n\nSee [docs\u002FLLM_TOOL_INTEGRATION.md](docs\u002FLLM_TOOL_INTEGRATION.md).\n\nThe browser workspace exposes the same evidence parameters: retrieval mode,\nyear range, edge type, method filter, graph depth, paper cap, and edge cap. It\nalso downloads the current evidence view as JSON, paper CSV, edge CSV, or\nMarkdown prompt context.\n\nThe workspace can use either the local SQLite graph or a hosted Intern Atlas API.\nChoose `Hosted API` in the sidebar, set the hosted base URL and optional API\nkey, then run the same evidence search. The browser calls the local FastAPI\nproxy at `\u002Fapi\u002Fv1\u002Fremote\u002F...`, so local frontends do not need to fight browser\nCORS rules.\n\n## Use The Hosted Intern Atlas API\n\nThe CLI can also call the hosted Intern Atlas API for larger graph evidence,\nidea generation, and idea evaluation.\n\n```bash\nintern-atlas remote health\n\nintern-atlas remote evidence \"efficient long-context attention\"\n\nintern-atlas remote context \"efficient long-context attention\"\n\nintern-atlas remote methods \"Transformer\"\n\nintern-atlas remote edges --method attention --limit 20\n\nintern-atlas remote ideas \"long-context efficient attention\" --use-llm\n\nintern-atlas remote eval \\\n  \"Use FlashAttention and LoRA for parameter-efficient vision transformer tuning.\"\n```\n\nHosted defaults can be configured once:\n\n```bash\nexport INTERN_ATLAS_REMOTE_BASE_URL=\"https:\u002F\u002Fyour-host.example.com\u002Fapi\"\nexport INTERN_ATLAS_API_KEY=\"YOUR_ATLAS_API_KEY\"\n```\n\nYou can also use the public site root as the base URL; the client normalizes it\nto the API root automatically:\n\n```bash\nexport INTERN_ATLAS_REMOTE_BASE_URL=\"https:\u002F\u002Fintern-atlas.opendatalab.org.cn\u002F\"\n```\n\nIf the public demo endpoint returns an upstream error, use your own deployed\nbase URL with the environment variable above or `--base-url`.\n\nThe local server also exposes hosted proxy endpoints:\n\n```bash\ncurl -X POST \"http:\u002F\u002F127.0.0.1:8000\u002Fapi\u002Fv1\u002Fremote\u002Fevidence\u002Fcontext\" \\\n  -H \"Content-Type: application\u002Fjson\" \\\n  -d '{\"query\":\"efficient attention\",\"mode\":\"deep\",\"base_url\":\"https:\u002F\u002Fintern-atlas.opendatalab.org.cn\u002F\"}'\n```\n\nFor direct website graph data from a customer laptop:\n\n```bash\nintern-atlas remote search \"FlashAttention\" --include-subgraph \\\n  --base-url \"https:\u002F\u002Fintern-atlas.opendatalab.org.cn\u002F\"\n\nintern-atlas remote query \"efficient attention\" --max-nodes 80 \\\n  --base-url \"https:\u002F\u002Fintern-atlas.opendatalab.org.cn\u002F\"\n\nintern-atlas remote chain \"attention\" --max-chains 5 \\\n  --base-url \"https:\u002F\u002Fintern-atlas.opendatalab.org.cn\u002F\"\n```\n\nSee [docs\u002FAPI_USAGE.md](docs\u002FAPI_USAGE.md) for endpoint details and Python\nclient examples.\n\n## Input Formats\n\nSupported input types:\n\n- PDF directory: `--pdf-dir .\u002Fpapers`\n- TXT file with repeated `Title:`, `Year:`, `Abstract:` blocks\n- JSON or JSONL with `paper_id`, `title`, `abstract`, `year`, `authors`, `venue`\n- CSV with at least `title,abstract`\n\nSee [docs\u002FLOCAL_GRAPH_BUILDER.md](docs\u002FLOCAL_GRAPH_BUILDER.md).\n\n## Security\n\nDo not commit `.env`, API keys, generated SQLite databases, or PDF corpora.\nThe `.gitignore` excludes those by default.\n","Intern Atlas 是一个用于从论文文件夹构建本地方法演化图的小型工具包。它支持PDF或标题\u002F摘要文件作为输入，并通过配置兼容OpenAI的聊天API来生成一个可由本地FastAPI服务查询的SQLite图数据库。该工具能够提取并标准化每篇论文中的方法名称，同时记录这些方法之间的引用关系（如引入、使用、扩展等）。适用于需要对某一领域内技术发展脉络进行梳理和分析的研究场景，特别是对于希望深入了解特定研究方向演变过程的学者或研究人员来说非常有用。","2026-06-11 04:02:42","CREATED_QUERY"]