Skip to content
Open agent platformBeta

Build with Kestrel.

Run real agent work without giving up control.

Give agents the durable runtime to carry work through—and keep the context, decisions, tools, and evidence you need to trust the result.

Choose your modelLocal or hostedOpen source runtime
Fix, test, hand off
Kestrel Desktop · Beta
Kestrel Desktop showing a completed checkout fix, regression coverage, and passing tests
The request, the result, and the verification stay together.

Choose your path

One platform. Three ways in.

Start locally, collaborate with a team, or bring the runtime into a product you already own.

Beta

01

Kestrel Desktop
Work locally with repositories, folders, research, data, and the artifacts Kestrel produces with you.
Beta

02

Kestrel One
Continue durable agent work across people, Projects, organizational Knowledge, Apps, and execution Environments.
0.7

03

Build with Kestrel
Use the runtime, CLI, SDK, framework adapters, and Execution Protocol inside products and operating systems you already own.

See Kestrel at work

The product is the proof.

The product surfaces are different because the work is different. The underlying model for durable agent execution stays consistent.

Kestrel Desktop
Beta
Kestrel Desktop showing a completed checkout fix, regression coverage, and passing tests

Kestrel Desktop · Beta

Work locally, with the project in front of you.

Open a repository, folder, research collection, or blank project. Kestrel can inspect the material, use approved tools, produce artifacts, and keep the session ready for your next return.

  • Local projects and model choice
  • Visible runtime and tool state
  • Durable sessions and artifacts
Download Desktop Beta

Durable work · visible control

Stay oriented while the work moves.

Kestrel treats agent execution as durable work rather than a disposable request. People remain part of the operating loop.

  1. 01

    Continue

    Sessions and context remain available when you close the window or return later.

  2. 02

    Supervise

    Review progress, answer questions, approve important actions, and redirect the work.

  3. 03

    Recover

    Interrupted or failed work leaves enough state and evidence to understand what happened.

  4. 04

    Inspect

    Runs preserve tool outcomes, checkpoints, artifacts, and explicit terminal results.

One runtime · two deployment forms

Run locally or remotely without changing the model of the work.

Desktop, CLI, Kestrel One, and server applications use the same execution architecture and explicit result contracts.

Local

Desktop · CLI · local SDK target

Local Core
Hosted

Kestrel One · remote SDK target

Runner service
Shared contracts

Kestrel runtime

The same execution model for sessions, runs, tools, persistence, cancellation, recovery, and terminal results.

Evidence and artifacts retained

Open platform

Build on explicit contracts.

Use Kestrel from the terminal or connect a trusted Node.js application to Local Core or a hosted runner service.

agent.ts
SDK 0.7
import { createAgent } from "@kestrel-agents/sdk";

const agent = createAgent({
  id: "project-agent",
  profileId: "reference",
  target: {
    kind: "remote",
    baseUrl: runnerUrl,
    authToken: runnerToken,
  },
});

const terminal = await agent.run({
  sessionId: "project-123",
  message: "Summarize the current project.",
}, context);
Beta

Help shape what Kestrel becomes.

Beta is a product stage, not a hidden program. We document the current state, ship in public, and use real feedback to improve the products.

Read release status

Desktop and Kestrel One are actively evolving Beta products.

Capabilities and availability may change as the products mature.

Release status and known limitations stay documented publicly.

Reproducible issues and product feedback directly inform development.

Start where the work is

Build something useful with Kestrel.

Work locally in Desktop, move together in Kestrel One, or bring the runtime into your own application.