An AI agent launcher with shared configuration, powered by agents-compat.
Work in progress — no usable code yet. See docs/design.md for the planned design.
Every AI coding agent (Claude Code, Cursor, Codex, Gemini CLI, OpenCode, ...) has its own config format, its own auth flow, and its own CLI conventions. agents-compat already solves shared instructions, skills, and MCP config by generating agent-specific files from open standards. agent builds on top of that: a single launcher command that keeps those generated configs current and then starts whichever agent CLI you asked for, with a consistent way to pick a provider, model, and variant across all of them.
- Switch agents without migrating config — instructions, skills, and MCP servers are defined once and synced to whichever agent you launch, so trying a new agent CLI doesn't mean copying rules and re-wiring MCP servers by hand
- One set of flags across every agent — pick provider, model, and variant the same way regardless of which underlying CLI ends up running, instead of relearning each tool's own flags/env vars
- Fail fast on incompatible setups — agent-locked providers (e.g. a Claude subscription) are rejected before anything launches, with a clear error instead of a confusing failure from the underlying CLI
- Never launch against stale config —
agents-compatsync runs before every launch, so generated files always reflect your current instructions/skills/MCP config
- Launch any agent supported by
agents-compat, syncing its generated config beforehand - Manage providers (add, remove, reauthenticate) independent of any single agent
- Select provider, model, variant, and agent via flags, validated against the current configuration
- Reject agent-locked providers (e.g. a Claude subscription) when used with an incompatible agent, with a clear error instead of a forwarded failure
- Pass provider-specific arguments straight through to the underlying agent CLI after
-- - Launch agents against local/cloud Ollama models via
ollama launch, opt-in via--ollamaor a global config default - Manage skills, hooks, and MCP servers, synced via
agents-compat - First-run wizard to enable/disable individual
agents-compatfeatures
Every agent supported by agents-compat is a planned launch target:
| Agent | Status |
|---|---|
| Claude Code | Planned |
| Gemini CLI | Planned |
| Cursor | Planned |
| OpenCode | Planned |
| Codex | Planned |