For agents

Set up your agent.

Connect your agent to models, compute and storage. Set its budget, choose its permissions and keep its work across sessions.

Access and funding

One agent. Its own access. Clear funding.

Identity

The principal identifies your agent and stays the same when its access key changes.

Access key

The credential lets your agent sign in. You can replace it or revoke access when needed.

Funding

The payer is the person or organization covering usage. The agent works within its assigned budget and permissions.

Getting started

From access to useful work.

01

Create an access key

An organization admin creates a service key for the agent, sets its budget and chooses which resources it can use.

02

Check the budget

Your agent can check its identity, who pays for its work and its remaining allocation before it starts.

03

Connect your agent

Use the API directly or connect a compatible agent through MCP. Both use the same account and permissions.

04

Save your progress

Agent Runtime keeps saved state across runs. You can replace an access key while keeping the same agent identity and saved work.

API reference

The details when you need them.

The API guide includes authentication, request examples and responses. OpenAPI provides the machine-readable contract, and MCP helps agents find and use available tools.

GET /v1/account/principalPOST /v1/account/service-credentialsPOST /v1/account/service-credentials/{credential_id}/rotatePOST /v1/agent-runtime/executePOST /v1/agent-runtime/stream

Stay in control

You set the budget and permissions.

Give each agent the resources it needs and review its usage as work progresses. Expanding its budget or permissions requires approval from the account owner or administrator.