Halfspan supervises the actions AI agents take in browsers and APIs. You define the operations an agent may perform, the limits it must follow, and the information it may access. Halfspan checks each proposed action and issues a permit when the required checks pass.
The browser use supervisor manages an agent's browser session. API integrations and the guardian proxy enforce permissions at your application's request boundary. Choose the path that fits where your agent runs and where you can enforce access.
Start exploring
How Halfspan works: follow an action from proposal to permission and execution.
Company setup: choose an integration and understand deployment and verification responsibilities.
API reference: inspect request and response schemas and examples.
API overview: understand authentication and service boundaries.
Recover a stopped run: distinguish refusal, clarification, missing authority, and operational failures.
Evaluate supervision: interpret decision quality, enforcement, and evaluation reports.
Verify evidence: check signed exports against a trusted public key.
Organization and agent responsibilities
Organizations define tasks and approve authority. A trusted runtime issues scoped service credentials and enforces grants. Agents propose actions against the current observation; they cannot mint credentials, expand grants, or approve their own transactions.
Your application remains responsible for authenticating accounts, authorizing access to resources, and executing business operations. Connect those checks to Halfspan and verify that protected requests cannot bypass them. Agents do not gain authority from discovery, a claimed provider name, or a bot score.
Read Data access and privacy to understand what information your integration processes and sends for review.