⚠️ Alpha — work in progress. This is an early, exploratory project: there is no functional release yet, the architecture is still being decided, and APIs may change. Not production-ready; do not depend on it.
A natural-language, multi-agent interface to AiiDA.
The goal is to let researchers create, run, inspect, and diagnose AiiDA workflows by describing what they want in plain language. Specialized agents call a controlled set of typed tools (via the Model Context Protocol) over a real AiiDA profile, grounded by retrieval over AiiDA documentation. Local models are a first-class target, not an afterthought.
This started as a Google Summer of Code 2026 project under NumFOCUS / AiiDA. It is exploratory — we're finding out what works first — but the goal is to grow it into a production-quality tool, not to stop at a prototype.
Where we stand at the beginning of July 2026:
aiida-agents-demo.mp4
Alpha / work in progress. Engineering scaffolding (tests, CI, linting, typing) is in place; feature work has not started. The architecture is being decided incrementally — read the design records before relying on anything here.
- Project plan:
docs/gsoc/project-timeline.md— phases and milestones. - Architecture decisions:
docs/adr/— the ADR log; start with ADR-01 (scaffolding). - Contributing / local setup:
CONTRIBUTING.md— theuv-based development workflow.
MIT. See LICENSE.