kotaemon
Use Kotaemon to build, deploy, configure, extend, and troubleshoot document QA, RAG, model-provider, ingestion, and Gradio app workflows.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- kotaemon
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use Kotaemon to build, deploy, configure, extend, and troubleshoot document QA, RAG, model-provider, ingestion, and Gradio app workflows.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/package-overview.md4.5 KB
- 参考文档references/repo-provenance.md2.6 KB
- 参考文档references/repo-routing-metadata.json500 B
- 参考文档references/troubleshooting.md4.2 KB
- 脚本scripts/check_install.py5.0 KB
03 SKILL.md 正文
Kotaemon
Use this repo skill when a task involves Kotaemon, ktem, document QA, RAG app deployment, document ingestion, model provider setup, GraphRAG, local-model configuration, custom Kotaemon components, or ktem app/index extensions.
Kotaemon is a Python workspace with:
kotaemon- core composable RAG library for components, documents, loaders, embeddings, LLMs, indexes, retrieval, ranking, reasoning, and QA.ktem- Gradio document-QA application with Resources, File Index, Chat, Settings, extension, MCP, and persistence layers.- root app entry points and deployment scripts for the user-facing document QA web UI.
Start Here
- For install or launch questions, first read
references/package-overview.md, then route tosub-skills/app-deployment/SKILL.md. - For code/API work, identify whether the task is ingestion, RAG pipeline composition, provider setup, or extension development and open the matching sub-skill.
- Run
scripts/check_install.pyfor a safe local diagnostic when a checkout or environment may be incomplete. It checks metadata/imports and optional package signals without starting the app or calling providers. - Read
references/troubleshooting.mdfor cross-cutting dependency, optional integration, credential, and configuration failures before changing code. - Read
references/repo-provenance.mdbefore deciding whether this skill is stale for a current checkout.
Route By Task
| Task signal | Read |
|---|---|
Docker/local install, uv sync, python app.py, Gradio launch, .env, flowsettings.py, app data, login, PDF.js, update scripts, Chroma migration |
sub-skills/app-deployment/SKILL.md |
| File upload ingestion, PDF/DOCX/HTML/XLSX/TXT/web readers, OCR/table parsing, splitters, document metadata validation | sub-skills/document-ingestion/SKILL.md |
Programmatic RAG, BaseComponent, Document, vector indexing/retrieval, reranking, QA/citations, prompt templates, reasoning chains |
sub-skills/rag-core/SKILL.md |
| OpenAI/Azure/Ollama/GGUF/local servers, embeddings, rerankers, web search, provider env keys, GraphRAG/NanoGraphRAG/LightRAG | sub-skills/model-providers/SKILL.md |
| Custom components, reasoning modes, file-index/retriever classes, pages/settings, pluggy extensions, project templates | sub-skills/extensions/SKILL.md |
Installation Context
- End users should prefer the documented Docker images or the repo's
uv sync --python 3.10path when using a checkout. - Manual developer setup commonly installs
libs/kotaemonandlibs/ktemeditable in the same environment. - The project has broad runtime dependencies: Gradio, LangChain provider integrations, LlamaIndex, ChromaDB/LanceDB, document parsers, provider SDKs, and optional local-model/GraphRAG packages. Install the smallest set needed for the selected workflow.
- Many integrations require external services, credentials, model downloads, or optional binaries. The bundled diagnostics are offline by default and should not be treated as provider connectivity tests.
Safe Diagnostics
python skills/kotaemon/scripts/check_install.py --repo-root <repo-root>
The root diagnostic reports Python version, discovered distributions, top-level import status, console script metadata, and whether common optional packages are installed. It does not run python app.py, start Gradio, call LLM APIs, download assets, or mutate data.
Runtime Boundaries
- This skill is self-contained for agent guidance; do not require future agents to open the original repo docs or scripts to use it.
- Bundled scripts are safe diagnostics or static validators. They intentionally avoid app startup, provider calls, model downloads, database migrations, and destructive writes.
- For repository development tasks, use this skill to choose focused evidence and tests, but verify against the current checkout before editing because public APIs and optional dependencies may drift.
References
references/package-overview.md- workspace package map, install variants, entry points, and dependency surfaces.references/troubleshooting.md- cross-cutting install/import/config/provider/app failures and routing guidance.references/repo-provenance.md- source snapshot and refresh baseline.references/repo-routing-metadata.json- structured metadata consumed by the managed repo-skills router during import.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).