ExecBound documentation
ExecBound sits between an AI agent and the systems it acts on. The agent proposes an action; ExecBound checks who the agent is, the trusted context and the policy, asks a person when a rule says to, and only then performs the action with a credential the agent never holds. Every decision is recorded, so it can be replayed and verified later.
To connect an agent to the hosted service, start with the quickstart; connect an agent host has the configuration for each host and the agent API. To run ExecBound yourself, install it on one host. What ExecBound claims about its own security, and the evidence for each claim, is on the trust center.
Start here
Sign in to the hosted console, or run ExecBound on one host or in your own cloud account.
Guides
Connect agents, write policy, route approvals and plug in the platforms you use.
- Connect an agent host
- Write a policy
- Users, roles and approval rules
- Review and approve requests
- Notifications and webhooks
- Govern Tines stories
- Discover agents in Entra
- The Arcade logic extension
- Monitor Tines stories
- Build an external executor
Concepts
One page each for the parts a decision is made from.
- Actions and the catalog
- Trusted context
- Policy and rulesets
- Approvals, impact limits and execution
- Providers and the credential boundary
- Evidence and replay
- The external execution checkpoint
- Integrations and the enforcement modes
- Security and the trust center
- Security signals
Technical reference
Interfaces in depth, the data model, deployment and recovery, for running and extending ExecBound.
- Interfaces in depth
- Surfaces in detail
- The data model
- The image and its processes
- Retention, backup and recovery
- Releases and verifying them
- Request references and external cases
- Glossary
- The identity mock
- Operations
- Audit event reference
- Error and refusal codes
- Command-line reference
- The executor protocol
Changelog
Every recorded change, newest first.
API reference
The agent API reference opens with authentication, requests, responses, errors and rate limits, then lists every route an agent calls; it is generated from the models the gateway validates with and checked against the routes it serves. The console API is the console's own interface, generated from its router, and marks whether a personal access token may call each operation.