Quickstart

Install via npm (Linux x64)

npm install -g @gair/codex-med

Currently only linux-x64 (glibc) prebuilt binaries are published. Other platforms can build from source — see Building from source.

Then run:

codex-med

Sign in

Inside the TUI, follow the prompts to sign in with your ChatGPT account or paste an API key. See docs/config.md for full configuration reference.

What's different from upstream Codex

This is a fork; everything in the upstream README about agents, sandboxing, MCP support, etc. still applies. The medical-fork-specific additions:

• Built-in biomedical lookup tools — fetch UniProt entries, GenBank records, PDB structures, and search UniProt directly from the agent loop
• Antibody training database tool — read-only SQL queries over a local SQLite database (training_ready_v1.sqlite); see docs/ for schema
• Vector knowledge search tool — search a Qdrant-backed medical vector knowledge collection directly from Codex with search_vector_knowledge; see docs/vector-knowledge-tool.md
• External model gateway integration — connect self-hosted prediction models (DLP-Affinity, structure prediction, ADMET, etc.) through the MCP gateway documented in docs/model-integration.md

Building from source

Requires Rust 1.95+ and Node 16+.

git clone https://github.com/openai/codex.git codex-med
cd codex-med/codex-rs
cargo build --release -p codex-cli --bin codex

The binary lands at codex-rs/target/release/codex. To package it as an npm tarball locally, see codex-cli/scripts/build_npm_package.py and scripts/stage_npm_packages.py.

Docs

• Model integration guide — how to plug self-hosted prediction models into the agent
• Vector knowledge tool — how to configure the built-in vector database search tool
• Qwen vector database deployment — how to deploy Qdrant, connect Qwen embedding/reranker services, import data, and use the database from Codex
• Configuration
• Contributing

This repository is licensed under the Apache-2.0 License.