AgentRun

AgentRun is an agent app that cuts repeat token costs and recovers interrupted runs by letting agents learn jobs, for teams running agents.

What it is good for

  • Cutting the token bill from agents repeating the same job
  • Resuming an AI agent run after the server crashes
  • Keeping track of what every agent is doing and spending
  • Requiring approval before an agent takes an action

At a glance

Who it is for
Setup difficulty
Technical ยท Guided command-line setup

Level 4 of 5. You copy and paste commands into a terminal by following the instructions. Comfortable territory for vibe coders.

Cost
Free and open source
Type
Agent app
Tasks
Works with
Claude Code, Codex, OpenCode, pi, OpenRouter, Docker

How to get started with AgentRun

  1. Install Node 24.22.3 or newer on macOS or Linux.
  2. Run npx @parcha/agentrun to start the server and open the console in your browser.
  3. Set a password in the console, then add a key under Settings โ†’ Model keys.
  4. Open Templates, apply KYB due diligence, and start a run with a company name and website.
  5. Open the agent's Learning tab to see what it kept for next time.

From the project's own documentation. If a step does not match what you see, the official pages have the latest instructions.

Before you try it

  • Requires Node 24.22.3 or newer on macOS or Linux.
  • Needs at least one model key, such as an OpenRouter, Anthropic or OpenAI key.
  • Without Docker running, agents can read and write files but get no shell.
  • The Docker sandbox image builds on first use and takes a few minutes.

Momentum

Rank this week
Not ranked yet
Best rank
None yet
GitHub stars
1,156

GitHub numbers as of Nov 16, 2026. Updated weekly.

685 938 1,191 Nov 13, 2026 Nov 16, 2026
GitHub stars for AgentRun: 720 on Nov 13, 2026, 1,156 on Nov 16, 2026. Source: GitHub.

On Hacker News

Questions

Is AgentRun free?
Yes. It is free and open source (Apache-2.0 license).
Do I need to know how to code to use AgentRun?
You do not need to write code, but setup means copying and pasting commands into a terminal by following its instructions.
What does AgentRun work with?
According to its own documentation, it works with Claude Code, Codex, OpenCode, pi, OpenRouter, and Docker.
How do I get started with AgentRun?
1. Install Node 24.22.3 or newer on macOS or Linux. 2. Run npx @parcha/agentrun to start the server and open the console in your browser. 3. Set a password in the console, then add a key under Settings โ†’ Model keys. 4. Open Templates, apply KYB due diligence, and start a run with a company name and website. 5. Open the agent's Learning tab to see what it kept for next time.
What should I know before trying AgentRun?
Requires Node 24.22.3 or newer on macOS or Linux. Needs at least one model key, such as an OpenRouter, Anthropic or OpenAI key. Without Docker running, agents can read and write files but get no shell. The Docker sandbox image builds on first use and takes a few minutes.
What is AgentRun good for?
Cutting the token bill from agents repeating the same job, resuming an AI agent run after the server crashes, keeping track of what every agent is doing and spending, and requiring approval before an agent takes an action.

Similar agents to AgentRun

  • OpenHands

    Self-hosted control center focused on coding agents, without the learning from past runs that AgentRun adds.

  • HiRoute

    Cuts cost by routing routine work to cheaper models, rather than having agents learn jobs.

  • Tucano

    Desktop app running a team of agents with approval before real-world actions, not a server platform.

  • Multica

    Assigns issues to coding agents like teammates, with no self-improvement or durable run recovery.

  • CCCC

    Coordinates several coding agents in one group chat, rather than adding learning and recovery.

  • OpenSRE

    OpenSRE is a command-line agent that investigates failed CI builds and pushes a proposed fix for engineering teams to review.

Also compared with AgentRun on these pages:

Last updated Nov 20, 2026. Tracked since Nov 13, 2026. Spotted an error? Tell us.