Your AI coding agent already reads code and writes code. DrBrain gives it a structured knowledge graph of academic papers — so it can search literature, trace causal chains, find research gaps, and infer new relationships through rule-based reasoning.
- 📚 Your paper library becomes a queryable knowledge graph with concept-level granularity.
- 🧩 Reasoning is symbol-driven: closure rules, confidence propagation, counterfactuals — not just embedding similarity.
- ⚡ Lightweight vectors for retrieval: semantically-complete tree nodes only, never arbitrary chunks.
- 🤖 Built for AI agents: every feature is accessible through the CLI that your agent already uses.
- Incremental by default — add one paper to an N-paper library and only that paper builds; closure scans its neighborhood; embeddings micro-adjust.
- 7 reasoning workflows — review, gap-analysis, impact, compare, frontier, lineage, paradigm.
- OKF export — emit the entire knowledge graph as an
OKF v0.1 markdown bundle that humans
and agents can read with
cat. - 27 agent skills following the open AgentSkills.io standard.
git clone https://github.com/UynajGI/DrBrain.git
cd DrBrain
uv sync && uv pip install -e .
drbrain setup # interactive wizard (bilingual EN / 中文)This creates ~/DrBrain/ as your library root
(%USERPROFILE%/DrBrain on Windows).
# Ingest → build → embed → closure (all incremental)
drbrain fetch "10.1038/nature14539" # grab a paper by DOI
drbrain build # 5-stage LLM extraction
drbrain embed # TransE graph embeddings
drbrain closure # rule-based inference
drbrain ask "What gaps remain in deep learning?"
# Or chain everything at once
drbrain pipeline --preset full
pipx install drbrainanduv tool install drbrainare coming in beta.
| Category | Feature | Details |
|---|---|---|
| Ingest | PDF → structured knowledge | MinerU parsing → 5-source metadata cross-validation (arXiv, CrossRef, S2, OpenAlex, DeepXiv) → LLM tree structuring |
| Build | 5-stage concept extraction (incremental) | Ontology extension → entity extraction (10-way concurrent) → relation extraction → coreference → iterative refinement |
| Query | BM25 + graph-enhanced search | Keyword search with multiplicative PageRank boost, directed graph traversal, hybrid ranking |
| Knowledge Graph | Rule-based closure (incremental) | 8+4 inference rules, t-norm transitive grounding, TransE embeddings for link prediction |
| Reasoning | Symbol-driven discovery | Causal chains, confidence propagation, counterfactual analysis, cross-domain isomorphism, hypothesis generation |
| Workflows | 7 structured reasoning pipelines | review, gap-analysis, impact, compare, frontier, lineage, paradigm |
| Sessions | Persistent reasoning context | DB-backed multi-turn sessions, build context injection, cross-invocation continuity |
| Analyze | Knowledge frontier reports | Research seeds, debate zones, technology cliffs, LLM executive summary |
| Citations | Multi-source expansion | Forward/backward citations, shared-reference analysis, citation verification |
| Export | BibTeX, RIS, Markdown + OKF | 4 citation styles (APA, Vancouver, Chicago, MLA) + OKF v0.1 markdown bundle |
| Import | Zotero, BibTeX, Endnote | Web API + local SQLite for Zotero, XML/RIS for Endnote |
| Genealogy | Concept lineage + descendants | Evolution trees, academic offspring, paradigm shift detection, cross-domain migration |
| Fetch | PDF acquisition from OA sources | 5-stage fallback (arXiv, OpenAlex, Unpaywall, direct DOI), proxy support |
| Federated Search | Local + arXiv | Cross-source search with ingested-status annotation |
| Patent Search | USPTO PPUBS + ODP | Free (PPUBS) or API-key (ODP) patent search |
| Pipeline | Step chaining (incremental) | Presets (full/quick/embed) + custom steps; --full forces rebuild |
| Audit | Data quality scan | 15 severity-graded rules, PDF pre-validation, ingest quality gates |
All commands
setup ingest fetch build embed closure query search ask
reason graph analyze evolve landscape frontier paradigm
citations export export-okf import translate session ws
pipeline repair enrich audit backup restore metrics
document fsearch patent-search proceedings explore check clean
Run drbrain --help for the full list, or see the
CLI Reference.
Install DrBrain skills so your coding agent can use them:
npx skills add https://github.com/UynajGI/DrBrain/skillsSkills follow the open AgentSkills.io standard and work with Claude Code, Codex, Cline, Cursor, Windsurf, Qwen Code, GitHub Copilot, and other AI coding tools.
drbrain setup walks you through the basics interactively (bilingual
EN / 中文):
- LLM API key (any litellm provider: OpenAI, Anthropic, Ollama, DeepSeek, …)
- MinerU token (optional; PyMuPDF fallback for PDF parsing)
- Semantic Scholar / CrossRef / OpenAlex API keys (optional; higher rate limits)
drbrain check verifies your environment end-to-end. See
Configuration for every setting.
| Getting Started | From install to first query |
| CLI Reference | All commands with examples |
| Architecture | System design and reasoning modules |
| Configuration | Every setting, default, and provider template |
| Workflows | Structured reasoning workflow guide |
| Sessions | Persistent Session Agent deep dive |
| Embedding | Local, openai-compat, and none providers |
| Troubleshooting | Common problems and recovery |
| Skills Reference | 27 agent skills and their CLI commands |
| Contributing | How to add commands, modules, and skills |
DrBrain's agent-first design is inspired by ScholarAIO — the pioneering "research infrastructure for AI agents." DrBrain takes a different technical path: symbol-driven knowledge graph reasoning with lightweight vector retrieval for semantically-complete nodes (vs full-text chunk embedding). Tree-structured retrieval is inspired by PageIndex and RAPTOR.
Contributions are welcome — bug reports, feature requests, docs, and code all help. See CONTRIBUTING.md for the full guide.
Quick start for contributors:
git clone https://github.com/UynajGI/DrBrain.git && cd DrBrain
uv sync && uv pip install -e .
pre-commit install # optional: auto-lint on commit
uv run pytest -m "not integration" # fast testsBefore opening a PR, make sure these pass:
uv run ruff check . && uv run ruff format --check .
uv run mypy src/drbrain
uv run pytest -m "not integration"We follow Conventional Commits
(feat:, fix:, docs:, refactor:, test:, chore:).
Have a question or idea? Open a discussion. Found a bug? File an issue.
MIT — see LICENSE.