THE RUNTIME BEHIND YOUR AGENTS

Give your agents
a place to work.

Persistent files and sessions for your product. A runtime designed to give each agent run its own place to work.

Built for developers putting agents inside their products.

Working local preview · Cloud execution in development

FROM THE LOCAL QUICKSTARTSYNTHETIC RUN

A file in. A result out.

A recorded run through the API and worker, using the installed SDK.

  1. Upload file
  2. Run completed
  3. Download result
Saved to the workspace
chalk-local-fixture.json

The local control-plane fixture completed.

simulated: true242 bytes downloaded

Your file stays.input.csv retained byte for byte.

A retry stays one run.Same intent, 1 run after retry.

The fixture checks the workflow. It does not analyze the CSV or run a customer agent. No cloud VM or model was used.

LESS INFRASTRUCTURE TO ASSEMBLE.One workspace per customerBring your agent codeCompute on demand

DESIGNED AROUND THE AGENT LIFECYCLE

The run ends.
The work carries on.

We’re building the pieces between your agent code and a dependable product experience.

01 / EXECUTE

A clean place to run.

Our cloud runtime is designed to give each active run a disposable virtual machine, with the tools and files your agent needs.

Cloud execution in development
02 / REMEMBER

Keep the useful parts.

Save committed files, messages and results to a customer’s workspace. Restore them when the next turn starts.

Durable files and sessions
03 / SHIP

Make it your product.

Connect through a typed API. Show progress, reconnect to a session and let customers cancel a run from your interface.

Your app. Your customer experience.

A SMALL API. ROOM TO BUILD.

From your product
to a persistent workspace.

Upload a file, start a run, follow its progress and download the saved result. The quickstart walks through the complete local workflow, including safe retries.

Read the local quickstart
your-app.tsSTART A SESSION
const workspace = await chalk.workspaces.create({
  customerId: "customer_001",
  name: "Monthly reporting",
});

const session = await chalk.sessions.create({
  workspaceId: workspace.id,
});

await chalk.runs.create(session.id, {
  input: "Find what changed this month.",
});

Opening steps shown. The quickstart adds stable retry keys, upload, completion checks and download. Local runs use a labeled synthetic fixture.

START SMALL. GROW WITH USAGE.

Compute when there’s
work to do.

Our GCP architecture starts with zero warm execution machines. Save the workspace, release the machine, and create a fresh one for the next run.

Stored files and platform services have ongoing costs. Public pricing will follow measured cloud usage during beta.

On-demand executionNo warm fleet by defaultUsage visibility

BUILD WITH US

Your next agent
starts here.

We’re building the private beta with developers shipping agent-powered products. Tell us what you have in mind.

This local preview saves your request on this machine. No email is sent.