The open source AI biomedical research agent.

[中文]

---

About Darwin

Darwin is a biomedical generative agent forked from and heavily modified on top of feynman. While feynman is a general-purpose AI research agent, Darwin specializes in systematic literature synthesis, clinical evidence grading, and reproducible in-silico protocols for the life sciences.

Key capabilities:

• Evidence-driven research — systematic reviews and meta-analyses following PRISMA guidelines
• Clinical evidence grading — hierarchy enforcement (RCTs > cohort > preprints) with inline retraction checks
• Reproducible protocols — wet-lab and computational pipelines with MIQE/CONSORT/ARRIVE checklists
• Bioinformatics integration — BLAST, pathway enrichment, GEO reanalysis, AlphaFold structure evaluation
• Standard nomenclature — HGNC gene symbols, RRIDs, UniProt IDs, ChEMBL IDs throughout

Built on the Pi agent runtime with biomedical expertise delivered as Pi skills — Markdown instruction files synced on startup.

---

Installation

From source (development):

git clone https://github.com/getcompanion-ai/darwin.git
cd darwin
npm install
npm run build

Then link the binary so darwin is available in your PATH. npm link sometimes fails in new terminals (nvm not loaded), so the recommended approach is a manual symlink:

mkdir -p ~/.local/bin
ln -s $(pwd)/bin/darwin.js ~/.local/bin/darwin
# Ensure ~/.local/bin is in PATH
export PATH="$HOME/.local/bin:$PATH"

To make PATH persistent across new terminals, add it to your shell config:

echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc

Or for zsh:

echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zshrc

Alternatively, if your nvm setup works across terminals, npm link also works:

npm link

To run without any linking, use node ./bin/darwin.js.

Requirements:

• Node.js >= 20 (tested up to v25)
• npm or yarn

Configure a model provider:

darwin setup

Choose from supported providers including DeepSeek, OpenAI, Anthropic, or local models via LM Studio / Ollama / vLLM.

---

What you type -> what happens

$ darwin "systematic review of GLP-1 agonists in heart failure"
-> Searches PubMed, clinical trials, produces a PRISMA-compliant review

$ darwin systematic-review "CRISPR base editing for sickle cell disease"
-> Multi-agent systematic review with evidence grading

$ darwin target-dossier "KRAS G12C"
-> Comprehensive target dossier with druggability assessment

$ darwin trial-tracker "donanemab Alzheimer's"
-> Clinical trial landscape with regulatory milestones

$ darwin retraction-sweep "amyloid hypothesis"
-> Scans for retractions and corrections in the field

$ darwin protocol "RNA-seq differential expression"
-> MIQE-compliant reproducible protocol

---

Workflows

Ask naturally or use slash commands as shortcuts.

Command	What it does
/deepresearch <topic>	Source-heavy multi-agent investigation with evidence hierarchy
/systematic-review <topic>	PRISMA-compliant systematic review or meta-analysis
/lit <topic>	Literature review from PubMed, bioRxiv, and primary sources
/target-dossier <gene/protein>	Comprehensive target validation and druggability assessment
/trial-tracker <drug/indication>	Clinical trial landscape and regulatory milestones
/retraction-sweep <topic>	Scan for retractions, corrections, and expressions of concern
/protocol <name>	Draft reproducible wet-lab or in-silico protocols
/biomarker-roc <marker+disease>	Biomarker performance evaluation with ROC analysis
/dosage-calc <drug+patient>	Drug dosage calculation with organ adjustment
/ic50-fit <compound+data>	Dose-response curve fitting with confidence intervals
/review <artifact>	Biomedical peer review with reporting checklist
/audit <item>	Evidence integrity audit
/compare <topic>	Source comparison matrix
/draft <topic>	Manuscript-style draft from research findings
/watch <topic>	Recurring literature monitoring
/outputs	Browse all biomedical artifacts

---

Agents

Six bundled biomedical agents, dispatched automatically.

• bio-researcher -- gather primary evidence across PubMed, clinical trials, bioRxiv, and regulatory filings
• clinical-researcher -- gather and appraise clinical evidence from RCTs, systematic reviews, and guidelines
• bioinformatician -- design and execute reproducible bioinformatics pipelines and in-silico analyses
• bio-reviewer -- simulated biomedical peer review with domain-specific checklists (CONSORT, ARRIVE, MIQE, PRISMA)
• bio-writer -- structured manuscripts, protocols, and evidence summaries
• evidence-verifier -- inline citations, source verification, retraction checks, and evidence hierarchy enforcement

---

Skills & Tools

• PubMed Search -- structured queries with MeSH terms for peer-reviewed literature
• bioRxiv / medRxiv Monitor -- track preprints before peer review
• Clinical Trial Check -- ClinicalTrials.gov and EU CTR search
• Retraction Check -- verify paper integrity before citation
• Gene Lookup -- HGNC, OMIM, ClinVar, UniProt queries
• Pathway Enrichment -- ORA and GSEA guidance
• AlphaFold Fetch -- retrieve and evaluate predicted structures
• Statistics Check -- audit statistical methods and p-hacking signs
• BLAST -- sequence alignment and ortholog search
• GEO Reanalysis -- reanalyze public datasets for validation
• Biomarker ROC -- diagnostic/prognostic performance evaluation
• Dosage Calc -- renal/hepatic adjustments with standard references
• IC50 Fit -- 4-parameter logistic regression with confidence intervals
• Web Search -- Exa, Perplexity, or Gemini API for current topics
• Session Search -- indexed recall across prior research sessions
• Preview -- browser and PDF export of generated artifacts

---

How it works

Built on Pi for the agent runtime, with biomedical capabilities delivered as Pi skills -- Markdown instruction files synced to ~/.darwin/agent/skills/ on startup.

Every output is source-grounded and respects the evidence hierarchy:

• Peer-reviewed systematic reviews and RCTs > cohort studies > preprints
• Retraction checks before citation
• In-silico results explicitly labeled, never presented as wet-lab confirmed
• Standard nomenclature: HGNC gene symbols, RRIDs, UniProt IDs, ChEMBL IDs