Developers
A decision in front of every tool call.
Keep building agents the way you do. Ask Lucidrail before a consequential call executes, handle the decision, and report the outcome.
Code on this page is illustrative and describes the integration model. It is not a published API reference. Talk to us for current SDK access.
Integration models
In-process or in-path.
SDK
Call authorize before tool execution inside your agent runtime. Best when you own the agent code.
agent runtime→authorize()→tool
Gateway
Route tool traffic through a governed proxy. Best for third-party agents you can’t modify.
any agent→lucidrail gateway→system API
Components
The building blocks.
Conceptual components. Availability of each interface is confirmed during onboarding.
Decision shape
Decisions explain themselves.
Each decision carries its status, the policy that produced it, the envelope level, and the evidence references — so your agent can respond sensibly and your logs stay useful.
// illustrative { "status": "approval_required", "envelope": "approval", "policy": "external-comms/contaminated-trajectory@v3", "reasons": ["task.provenance = contaminated"], "approver": "dana.whitfield", "evidence": ["inbox/msg_8812.pdf"], "expires": "2026-09-30T18:02:07Z" }