Overnight jobs, reviewed over coffee.
Dependency upgrades, flaky tests, migrations, docs nobody wrote. Queue them up at the end of the day and review the branches in the morning.
Pre-launch. Accounts are open; the API and checkout open soon.
How it runs.
01
Write the jobs down.
One issue or prompt each: upgrade this dependency, fix that flaky test, document this module.
02
Start the agents as you log off.
Each works in its own branch or worktree and runs the tests as it goes.
03
Review in the morning.
Read the diffs, merge what passes, send back what doesn’t.
Why lanes fit it.
The night is included.
Eight hours of agents working costs nothing on top of the plan.
Retries don’t cost more.
An agent that reruns the tests five times holds its lanes longer, not your money.
Queued, not capped.
When every lane is busy, the next job waits for one to free up instead of failing.
Works with what you use.
OpenCode
opencode.json
{ "$schema": "https://opencode.ai/config.json", "provider": { "voidstone": { "npm": "@ai-sdk/openai-compatible", "name": "Voidstone", "options": { "baseURL": "https://api.voidstone.net/v1", "apiKey": "{env:VOIDSTONE_API_KEY}" }, "models": { "kimi-k3": {}, "glm-5.3": {}, "deepseek-v4.1-flash": {} } } }}