RENKIN

Local retrosynthesis planning and route auditing in Rust.

Docs · Playground · 日本語 · 中文

RENKIN has two complementary jobs:

• Plan: search retrosynthetic routes from target SMILES to declared building blocks.
• Audit: apply the same deterministic structural, stock, forward-replay, and provenance checks to routes from RENKIN, AiZynthFinder, Syntheseus, or SynPlanner.

Current release: v1.0.10. Core chemistry is pure Rust; the project ships a CLI, Rust crate, Python package, MCP server, and browser WebAssembly module. The default planner has 24 hand-crafted rules. Repository-backed runs use the 402-compound stock file; WASM and installations without that file use a compiled-in 152-compound fallback. Supply stock explicitly when it matters.

Install

pip install renkin
cargo add renkin
npm install renkin

Start in two commands

Plan aspirin from the CLI:

renkin --target "CC(=O)Oc1ccccc1C(=O)O" --depth 5 --beam-width 100

Audit an exported route locally:

renkin audit-route route.json --format auto --output json

Python uses the same engine:

import json, renkin
routes = json.loads(renkin.find_routes("CC(=O)Oc1ccccc1C(=O)O", depth=5))
report = json.loads(renkin.audit_route(open("route.json").read(), format="auto"))

Choose an interface

| Need | Start here | | --- | --- | | Browser-only exploration or audit | Playground · WASM API | | Application integration | Python · Rust | | Local agent workflow | MCP guide | | Private stock or route evidence | Audit guide · Policy guide |

renkin capabilities, renkin.capabilities(), WASM capabilities(), and modern MCP discovery expose each surface's effective limits. Private inputs stay local unless an operator explicitly exports them.

Scope and benchmark boundary

An audit result is evidence, not a claim that a synthesis will succeed in the laboratory. Stock identity is exact standardized canonical-SMILES membership; model output and route imports are never accepted without validation.

The registered Phase 55 shared-stock TEST found strict routes for RENKIN on 481/690 targets (69.71%) and AiZynthFinder 4.4.1 on 32/690 (4.64%), under its pinned cohort, stock, assets, and budgets. It does not establish universal planner superiority, laboratory viability, or whole-cohort speed superiority. Read the result record before reusing the figure.

Development

cargo test --workspace
cargo clippy --workspace --all-targets -- -D warnings
cargo fmt --all -- --check

For architecture and contribution rules, see AGENTS.md, CONTRIBUTING.md, and the current ROADMAP.md. Full release history remains in CHANGELOG.md.

License

MIT. See LICENSE.