Model Picker
Describe a task and the Model Picker ranks the models that fit it, with the evidence behind each pick, so you can see why one wins.
How it works
From a task to done, through one router.
You describe the work. Eyrie picks who does it, runs it, and sends the result through review before it counts as finished.
- 1
Task
Ask in plain words, or pull the next task from the shared queue.
- 2
Router
Eyrie weighs the job and picks the best model and harness for it.
- 3
Agent or harness
The work runs in Codex CLI, Claude Code, ZeroClaw, or a Cursor cloud agent.
- 4
Review gate
A reviewer checks the result before anything lands.
- ✓
Done
Approved work is merged and the task closes.
Inside Eyrie
Three things you'll use every day.
Commander
Ask in plain words. The Commander routes the work to Codex, Claude Code, ZeroClaw, or Cursor agents, and keeps track of it.
One-command install
Run one command, paste one model key, and you're running. A good key works right away, with no restart.
Install command coming soon.
Supported harnesses
Bring the agents you already use.
Eyrie doesn't lock you into one vendor. It sends each task to the harness that suits it.
Codex CLI
Terminal coding agent
Claude Code
Agentic coding in your repo
ZeroClaw
Lightweight agent runtime
Cursor cloud agents
Background agents in the cloud
The vision
A software factory, run by agents.
Specialized AI agents work a shared, claimable queue of tasks behind a reviewer gate. Eyrie is the routing and execution layer underneath it.
- Shared queue. Every task lives in one place, and any agent suited to it can claim it.
- Specialists. Each agent does the kind of work it's best at.
- Reviewer gate. Nothing ships until it has been reviewed.
Demo
See Eyrie run.
Placeholder
Demo video coming soonThe new first-run walkthrough is being filmed.