🤖 Orchestrate a multi-vendor coding fleet
Stack — Trigger: chat / agent API · Models: per-CLI (Claude Code, Codex, aider) · Tools: runner process surface + coding CLIs · Mode: planner + parallel members · Output: edits in one workspace
Who it’s for — engineers who want more than one coding agent on a task, each on its strongest model. What it does — one planner agent breaks a task down and dispatches subtasks to several coding-agent CLIs running on your machine — each with its own model and isolated config — all editing the same repo, while you watch their status.
Builds on Give an agent a machine: driving coding CLIs is one capability the runner offers once a machine is exposed.
Build it
-
Start a runner on the machine with your repo:
npx @almyty/auth login npx @almyty/runner start --name my-laptopIt registers and reports the coding CLIs it finds on your PATH. Code and credentials stay on the box; only instructions cross the wire.
-
Create a planner agent whose instruction is to decompose the task and dispatch subtasks to the specialists.
-
Launch the CLIs as members — each unattended, each with its own model, isolated config, and headless auth.
-
Let them work the same workspace — the planner assigns files/subtasks; you watch each member’s status (busy / idle / awaiting input).
-
Review the result — the members edit one shared workspace; you review the diff as usual.
Make it yours
- Put a different model behind each CLI so the fleet is genuinely multi-vendor.
- Pair with the PR review bot to check what the fleet produced.
- The coding CLIs authenticate to their own providers with keys you supply, so those model calls leave from your machine.
Uses — DevOps · Runner · Coding agents