BACKED BY Y COMBINATOR

The harness behind grep.ai

We are open-sourcing its key pieces as pi extensions. Install the ones you need.

Set up AgentRun with your agent

Use the native extension for Pi, or connect another harness through adapters. These instructions use AgentRun 0.1.0-beta.27.

What you need

Jev + your agent
A Jev API key, your authenticated agent runtime, and the tools the workflow calls. Reuse your existing server-side key. Get a Jev API key ↗
Runtime
Node 22.19+ and npm. Use release 0.1.0-beta.27 for both packages and example source. Other harnesses use the @parcha/agentrun-dsl and @parcha/agentrun-jev packages. TypeScript 5.4+ if you use TypeScript.

Using Pi? Install the native AgentRun extension. Other harnesses connect through adapters. Connect your agent →

Use with Pi

Less planning on repeat runs.

SUPPORT REQUESTS1 / 3

“How do I reset my password?”

Agent alone

Plans each request

With AgentRun

Reuses the same steps

Define once. Reuse on every request.

You set the limits and the review fallback.
OPEN-SOURCE PIECES

What your agent can do

Each part is its own open-source package. Install one, or all of them.

Available now

Sees

Ask about a PDF, a scan, a photo, a video or a recording. The agent cites the page, second or spot behind each answer, and Jev checks what it sees.

See examples @parcha/pi-vision
Available now

Browses

Your agent gets its own browser, and you can take over and hand it back. Every page it reads is filed as evidence.

Read more @parcha/pi-browser
Available now

Runs anywhere

The agent is the workspace: its runs live on a disk you can mount on any Linux host or container. Jev places its tool calls on Daytona and Modal sandboxes, and they can also run as serverless calls on an Archil disk.

Read more @parcha/pi-durable-disk @parcha/pi-transport
Coming soon

Gets better at its job

As it works, it notes what would help its next run, and the host keeps those notes in the agent's own git repository. A curator merges the good ones to main, and git revert undoes a bad one. It can save a job as a workflow, and once one is proven, the host can run it first.

@parcha/pi-agent-learn @parcha/agentrun-dsl @parcha/pi-git-remote
Available now

You host it

Your agents live in one folder: each one's revisions, prompt or workflow, and every run, which names the revision it ran. Copy the folder and they all come with it. pi-durable-agent serve opens them on your own machine.

@parcha/pi-durable-agent
WORKFLOWS

Workflows

Write a job as a typed workflow, and run it with the agent you already use.

How it works

Define it once.
Run it with your agent.

Give repeatable work an explicit procedure. Jev makes typed decisions, your agents do the work, and code controls what happens next.

  1. Define the steps

    Write the workflow in TypeScript or JSON. Set its inputs, branches, checks, and stopping conditions.

  2. Connect your agents

    Supply adapters for Jev, your agents, and tools. Your harness keeps control of models, permissions, and budgets.

  3. Test. Then repeat.

    Check decisions and complete runs against known cases. Expose the tested workflow as a tool your agent can call.

Example / Weekly release notes

Illustrative workflow
  1. Tool

    Load merged PRs

    #412 · #419 · #421

  2. Jev / sift

    Select user-facing changes

    Keep #412 and #419

  3. Your agent

    Draft release notes

    Write from selected changes

  4. Code

    Check source IDs

    Every reference must exist

OutputA sourced draft, ready for review.

See the workflow guide

The workflow defines the procedure. Your harness runs it.

USE CASES

Where a workflow helps

Start with a task you run often: triage requests, review documents, or research a set of questions.

Select a card to compare the approaches.
THE LANGUAGE

What goes into a workflow

AgentRun is a domain-specific language (DSL) for defining these steps in TypeScript or JSON. The runtime calls your adapters and validates outputs against the schemas you declare.

TYPED ANSWERS

Jev decisions

Return an answer and its probability for the next step to use.

judge
Ask typed questions about the evidence.
route
Choose a branch by meaning. Set a fallback for uncertainty.
pick
Select one item, or none, and retain the probabilities.
sift
Evaluate a collection and keep the items that qualify.
TASK EXECUTION

Agents and code

Use agents for judgment and code for exact rules.

agent
Run an agent and validate its structured result.
decide
Ask your agent for a structured decision.
extract
Ask your agent to return specific fields.
code
Transform state with trusted JavaScript.
call
Invoke a host tool, executor, or shell with a deadline.
ORDER AND LIMITS

Control flow

Choose which steps run and how often.

chain
Run steps in order.
map
Process each item with bounded concurrency.
parallel
Run independent branches at the same time.
loop
Repeat until a condition is met, within an iteration limit.
workflow
Reuse a child workflow with its own input and output.
WORKFLOW OUTPUT

Results and handoffs

Return a result or stop with a reason for your host.

escalate
Stop with a reason and the current state for your host.
report
Generate a final report through your agent adapter.
artifact
Declare a final report or file for your host to deliver.

Inspect a run

Inspect execution events, Jev answers, and selected branches. Your host can record adapter inputs and outputs.

Test a decision

Save a node’s inputs and compare models or rubrics without repeating the full workflow.

PI EXTENSION

Build and run workflows in Pi

Describe a task in Pi. The AgentRun extension builds a workflow, inspects its graph, and runs its steps with your current Pi model.

Use your existing Pi model access. For Jev decisions, configure TYPESAFE_API_KEY in Pi’s environment. Set up Jev

pi × agentrun0.1.0-beta.27

Requires Node 22.19+ and Pi (tested with 1.3.0). Install in your project:

pi install npm:@parcha/agentrun-pi@0.1.0-beta.27 -l
pi --offline
THEN, INSIDE PI
/agentrun Find the main entry points in this repository and summarize how requests reach them.

Review Pi’s project-trust prompt on first launch. The extension uses your configured model; --offline only skips startup downloads.

Build your first workflow

Copy the setup instructions into your coding agent, or follow the quickstart.

Set up AgentRun with your agent

Use the native extension for Pi, or connect another harness through adapters. These instructions use AgentRun 0.1.0-beta.27.

What you need

Jev + your agent
A Jev API key, your authenticated agent runtime, and the tools the workflow calls. Reuse your existing server-side key. Get a Jev API key ↗
Runtime
Node 22.19+ and npm. Use release 0.1.0-beta.27 for both packages and example source. Other harnesses use the @parcha/agentrun-dsl and @parcha/agentrun-jev packages. TypeScript 5.4+ if you use TypeScript.

Using Pi? Install the native AgentRun extension. Other harnesses connect through adapters. Connect your agent →