# AgentFEM

> AI-native finite-element computing for humans and agents.

AgentFEM keeps finite-element models readable, structured, inspectable, and
operable through ordinary Python and a machine-readable CLI.

## Start

- [Human quick start](get_started/)
- [AI agent quick start](agents/)
- [Engineering guides](guide/)
- [Examples](examples/)
- [Python API](reference/api/)
- [Machine manifest](agentfem.json)
- [Installation and platform guide](getting_started/)
- [Windows short onboarding](windows_agent_installation/)

Installation is agent-assisted: inspect the OS, retain the user's mirror
choice, reuse existing environments, and check installation readiness.
Stop after reporting version, environment and readiness. Offer a first
example; create or run it only when requested. No automatic follow-on tasks.
No repository clone is required. On Windows follow the short onboarding
route; after one failed stage show one manual action, not a retry loop.
Read installation.onboarding in agentfem.json for the shared contract.

## Everyday use

For an existing project: read its instructions and case, check, run, inspect.
For a result-only request: show the existing result without a new solve.
Reuse a healthy environment; doctor is for installation, environment changes
or runtime failures. Read only the relevant physics/API references.
Use installed capabilities when web documentation describes a newer version.
Read task_guidance in agentfem.json for task-specific routing.

## Canonical sources

- [Repository](https://github.com/haoming-luo/agentfem)
- [Scientific knowledge catalog](https://raw.githubusercontent.com/haoming-luo/agentfem/main/src/agentfem/knowledge/catalog.json)
- [AgentFEM skill](agents/skill/)

## Result meaning

A completed run supplies a computed result. Provenance verification checks
the saved record; engineering acceptance uses relevant convergence, balance
and reference evidence. Match the checks to the user's intended use.