{"name":"AgentRun: Manage Agents and Revisioned Runs","description":"AgentRun is a self-hostable platform for managing agents, their revisions, and runs, built on Pi’s durable runtime. Its workflow DSL supports typed decisions and agent tasks, while the server provides a single API and interface for inspecting runs.","github":"https://github.com/Parcha-ai/agentrun","url":"https://osrepos.com/repo/parcha-ai-agentrun","source":"osrepos.com","sourceDescription":"This repository profile is provided by osrepos.com, an open source repository discovery platform.","repositoryProfile":"https://osrepos.com/repo/parcha-ai-agentrun","generatedFor":"open source discovery and AI-assisted research","markdown":"https://osrepos.com/repo/parcha-ai-agentrun.md","json":"https://osrepos.com/repo/parcha-ai-agentrun.json","topics":["typescript","ai-agents","self-hosted","workflow-automation","agent-orchestration","agent-framework","agent-workflows"],"keywords":["typescript","ai-agents","self-hosted","workflow-automation","agent-orchestration","agent-framework","agent-workflows"],"stars":null,"summary":"AgentRun is a self-hostable platform for managing agents, their revisions, and runs, built on Pi’s durable runtime. Its workflow DSL supports typed decisions and agent tasks, while the server provides a single API and interface for inspecting runs.","content":"## Overview\n\nAgentRun is a platform for creating and changing agents, then running them with a record of the exact revision used. Its server exposes agent revisions and runs through one API and page. The platform runs on Pi by default, with Claude Agent SDK and Codex available as named harnesses when supported by the server. AgentRun builds on Pi’s durable runtime, including pi-durable and chord.\n\nThe accompanying TypeScript workflow DSL describes repeatable steps, typed decisions, and calls to agents. This is aimed at teams that want workflows they can inspect, test, and run through an agent platform rather than keeping the process only in application code. You can self-host the server or use the packages individually in a Pi-based application.\n\n## Key Features\n\n- Store agents, revisions, and runs together, with each run tied to the revision it used.\n- Host the server in a Docker container, with its data kept in a mounted directory.\n- Define workflows as JSON or with a TypeScript builder and Zod contracts.\n- Combine tool calls, Jev typed decisions, and agent tasks, with branching, parallel work, and bounded loops.\n- Run scripted demos and tests without model keys before connecting live model services.\n- Use the Pi extension to build, inspect, and run workflows from Pi.\n- Add separate packages for browser access, document and video vision, tool transport, and durable agent storage.\n\n## Use Cases\n\n- **Support teams** can search for an answer, check it against defined criteria, ask an agent to investigate when needed, and route unresolved cases for review.\n- **Developers integrating agents into an application** can put repeatable steps and decision thresholds in a workflow, then connect their existing tools and agent through adapters.\n- **Teams reviewing agent behavior** can inspect which agent revision ran and examine the corresponding run record.\n- **Pi users** can create and execute workflows from the Pi coding agent, including scripted runs that do not require model calls.\n\n## Our Take\n\nAgentRun is an unusually broad, actively changing agent platform and workflow toolkit, but its newness and dependence on one primary contributor make it a higher-uncertainty choice for production adoption.\n\n**Good fit if:**\n- You want to self-host a platform that records agent revisions and runs, and are comfortable operating a beta-stage project.\n- Your workflow needs explicit typed decisions, review paths, or agent tasks rather than only a fixed sequence of application functions.\n- You use Pi and want its durable runtime and workflow packages as building blocks.\n\n**Look elsewhere if:**\n- You need a mature project with demonstrated release history and broad contributor ownership.\n- Your agents must use a harness other than Pi, Claude Agent SDK, or Codex, or you need a harness package that is already published rather than held.\n- You want typed decisions to guarantee factual correctness. The project explicitly notes that schemas and decision thresholds do not establish truth.\n\n## Project Health\n\n| Signal | Value |\n|---|---|\n| Status | **New**: less than six months old and changing fast |\n| Last commit | 2026-11-02 (today) |\n| Commits, last 90 days | 100+ |\n| Releases, last 12 months | 1 (latest gallery, 2026-10-20, pre-release) |\n| Contributors | 3 (top contributor: 100% of commits) |\n| Issues closed, last 90 days | 0+ |\n| Pull requests merged, last 90 days | 28 (typically merged in 0 days) |\n| Project age | 40 days |\n\nChecked on 2026-11-02 with the GitHub API.\n\n## Project Facts\n\n- Language: TypeScript\n- License: Apache-2.0\n- Stars: 423\n- Forks: 34\n- Archived: no\n\n## What You Need\n\nDetected in the repository:\n\n- Node.js >=22.19.0 (from package.json)\n- A test suite and automated checks on GitHub Actions\n\n## Getting Started\n\nFor a scripted DSL demo with no API key, use Node.js 22.19 or newer:\n\n```sh\nnpm install @parcha/agentrun-dsl@beta\nnpx @parcha/agentrun-dsl@beta demo\n```\n\nTo try the self-hosted platform, see the [repository README](https://github.com/Parcha-ai/agentrun) and [AgentRun Server guide](https://github.com/Parcha-ai/agentrun/tree/main/packages/serve).\n\n## License in Plain Words\n\n**Apache-2.0** (permissive).\n\n- Commercial use: yes\n- Modify and redistribute: yes\n- You must keep: the license, the NOTICE file and a note of your changes\n- Share your changes: no\n- Includes an explicit patent grant from the contributors\n\nA summary, not legal advice: the LICENSE file is what applies.\n\n## Alternatives\n\n- [swarmclaw](https://osrepos.com/repo/swarmclawai-swarmclaw): SwarmClaw focuses on coordinating persistent, tool-using agents with schedules and delegation, while AgentRun centers on versioned agents and inspectable workflow runs.\n\n| Project | Language | License | Stars | Status |\n|---|---|---|---|---|\n| **agentrun** | TypeScript | Apache-2.0 | 423 | New |\n| [swarmclaw](https://osrepos.com/repo/swarmclawai-swarmclaw) | TypeScript | MIT | 697 | Quiet |\n\n## Considerations\n\n- The project is very new and changing quickly. Recent activity is high, but most contributions come from one person; the measured data also shows no issues closed in the recent period, despite merged pull requests.\n- Published packages are distributed under the `beta` tag, and the latest listed release is a prerelease. Several packages and non-Pi harness adapters are held back from npm.\n- The server and workflow tools still require you to supply model access for live runs. The scripted examples avoid that requirement, but do not demonstrate live model behavior.\n- Workflow code nodes run with process privileges. The README warns that hosts must provide a controlled sandbox when workflow authors are untrusted.\n- Self-hosting with persistent run history requires durable storage. The README notes that Render's free instance has no disk and loses data when it sleeps or restarts.","metrics":{"detailViews":5,"githubClicks":2},"dates":{"published":null,"modified":"2026-11-02T16:37:27.000Z"}}