AgentRun
Open-source layer from Parcha that runs your Claude Code, Codex, OpenCode or pi agents, keeps what they learn and resumes runs after a crash.
- AI tool
- Free
Pricing
Free and open source (Apache 2.0). You run it yourself and bring your own model keys, for example one OpenRouter key.
Overview
AgentRun is an open-source platform from Parcha that sits on top of the agent harness you already use, such as Claude Code, Codex, OpenCode or pi. After each run it keeps notes on what the agent learned, so the next run on the same job can skip rediscovering the same APIs and plans and use fewer tokens, sometimes on a cheaper model. Runs are durable: if the server stops mid-run, the run carries on when it starts again. Each agent can get a sandbox, a Linux desktop, an Android phone or a browser, and one console shows approvals, spend caps, failures and cost for every agent. You start it with one npx command on macOS or Linux, or deploy it to Render. You get a console for running your agents that keeps their lessons and their costs in one place.
Key features
- Works with Claude Code, Codex, OpenCode and pi
- Agents keep notes on what they learned
- Durable runs that survive restarts
- Sandbox, Linux desktop, Android phone or browser per agent
- Approvals and spend caps
- One console for runs, failures and cost
- Open source under Apache 2.0
Use cases
Show all 10Show fewer
Alternatives
Last updated 2026-11-13