Security for AI agents
Let agents work.
Set the rules.
Steer AI agents with project-specific security policies. Monitor policy lookups, see which rules matched, and review the decision bonsai3 returned.
Connect with MCP using a project-scoped token.
“Build an account export endpoint”
Authorize exports on the server. Scope records to the authenticated tenant.
How it works
From intent to decision.
Set the policies once, make them available to the agents on that project, and review the decisions they produce.
- 01 / DEFINE
Write project policies.
Combine built-in secure coding guidance with your own rules. Choose what to permit, what to block, and when to steer.
See policy examples - 02 / CONNECT
Bring your agent.
Give agents a project-scoped MCP token. They can look up applicable policies and get guidance for the task at hand.
Connect an MCP client - 03 / REVIEW
See what matched.
Review decisions by project and trace outcomes back to the policy versions that produced them.
Agent security, by project