Based on AgentRun release · Synced .
Packages and releases
Start with the platform quickstart. The AgentRun launcher installs the server it was released with; you do not need to assemble extensions to run the platform.
The table below comes from the current repository snapshot. Publication was checked against npm on 2026-11-10. Package versions are independent: the launcher is 0.4, while the server and other published building blocks use 0.1.0-beta.32. A matching version number in source does not mean that every change on main is published.
| Package | Purpose | Version in source | Availability |
|---|---|---|---|
@parcha/agentrun |
Start AgentRun on this machine with one command: npx @parcha/agentrun | 0.4.0 | Published |
@parcha/agentrun-serve |
The Agentrun v2 API on your own machine: an agent store's agents, revisions and runs, served on loopback. | 0.1.0-beta.32 | Published |
@parcha/agentrun-computer |
The open platform's computer protocol: a daemon in every computer, and one provisioner per vendor, starting with Docker. | 0.1.0-beta.32 | Published |
@parcha/agentrun-client |
A typed TypeScript client for a self-hosted AgentRun server: a method per operation of its API, typed from its OpenAPI document, with waiting for a run and following its timeline. No dependencies; fetch only. | 0.1.0-beta.32 | Published |
@parcha/agentrun-decisions |
Third-party decision models as DecisionProviders for the AgentRun DSL: Parallel Decisions (typed questions grounded in live web evidence) and OpenRouter's Decisions models (any of them by name), behind the DSL's runJudge seam. | 0.1.0-beta.32 | Published |
@parcha/agentrun-dsl |
Composable, typed workflows for agents | 0.1.0-beta.32 | Published |
@parcha/agentrun-harness-claude |
The Claude Agent SDK as an AgentRun harness: the host's tools as its only tools, its model behind a loopback proxy, durable in AgentRun's own store. | 0.1.0-beta.32 | Published |
@parcha/agentrun-harness-codex |
Codex as an AgentRun harness: a tool-less Codex app-server behind a loopback model proxy, durable in AgentRun's own store, with the host's tools as its only tools. | 0.1.0-beta.32 | Published |
@parcha/agentrun-jev |
Optional Jev system one decisions for AgentRun workflows | 0.1.0-beta.32 | Published |
@parcha/agentrun-kernel |
Agent harnesses other than Pi, starting with Codex, behind one seam: a DSL node's runNode and an AgentRun run on a harness adapter, durable in AgentRun's own store. | 0.1.0-beta.32 | Published |
@parcha/agentrun-pi |
Pi extension for building, running and evaluating AgentRun workflows | 0.1.0-beta.32 | Published |
@parcha/pi-browser |
Your agent gets its own browser, and you can take over and hand it back. | 0.1.0-beta.32 | Published |
@parcha/pi-durable-agent |
An agent you can carry: its definition, sign-ins and runs, as one directory of pi-durable stores. | 0.1.0-beta.32 | Published |
@parcha/pi-durable-disk |
Run a pi-durable agent from a directory or a network disk, so any process or host can open the run and carry on where the last one released it. | 0.1.0-beta.32 | Published |
@parcha/pi-transport |
A pi agent's tools on the machine each step needs: this machine, Docker, or a Daytona or Modal sandbox, behind one Box interface; Jev places each step. | 0.1.0-beta.32 | Published |
Choose an installation
- Run the platform: use
npx @parcha/agentrun@0.4and follow the quickstart. - Call the API from code:
npm i @parcha/agentrun-clientorpip install parcha-agentrun. See Use the API. - Develop against the latest repository: use a checkout of main and follow the source setup. Main can include changes awaiting a release.
- Embed a building block: install its published version from the table and read that package's reference. A source-only package needs a source checkout; an npm install command will not work.
Server image
The published server image is ghcr.io/parcha-ai/agentrun-server:v0.1.0-beta.32; there is no latest image tag. The beta tag currently resolves to that same release. The current main branch adds support for Docker boxes beside a containerized server that is not in that published image. Use the local launcher on a Docker-equipped host for the supported local shell setup, or follow the current source deployment instructions when building from main.