8. AI providers
Aefos is "bring your own CLI": it ships no model and manages no credentials. You install and log in to the AI CLI of your choice, and Aefos connects it to your project. Four providers are supported.
Overview
| Provider | CLI | What you need |
|---|---|---|
| Claude Code | Anthropic | Install the CLI and log in to your Anthropic account |
| Codex | OpenAI | Install the CLI and run codex login to use the model |
| GitHub Copilot CLI | GitHub | Install the CLI and log in to GitHub Copilot |
| Gemini | Install/have the Gemini CLI available and log in |
Claude Code is the reference provider (the MVP target). The others are fully supported.
How to choose the provider
In Tools → Options → Aefos, select the provider/executor and the model. The model is remembered per provider (no leaking from one to another).
Connecting MCP to each CLI
Aefos points all CLIs to a single global MCP configuration
(%APPDATA%\Aefos\aefos-mcp.json). Each CLI is wired in its own way:
- Claude Code — receives the global config strictly (ignores any project
.mcp.json). - Codex — the
aefosserver config is injected per invocation (no config files touched). Remember Codex needscodex loginto run the model (the MCP connects regardless). - GitHub Copilot CLI — the global config is injected and tools are allowed for non-interactive mode.
- Gemini — the
aefosserver is merged into Gemini's settings (~/.gemini/settings.json) and enabled by name.
You don't need to do this wiring by hand — Aefos handles it. To verify that everything is fine, use Test MCP in Options → Aefos (see Configuration).
What if the CLI is not installed?
- The installer detects
claudeon your machine and, if missing, points to the official source. - At runtime, if you pick a provider whose CLI is not installed/logged in, the dispatch has nowhere to go — install/log in to the CLI and try again.
➡️ Next: Configuration