Developer platform
Adding control should feel like adding observability
Agent Trust Cloud must be easy for internal platform teams and software vendors to embed. The developer platform should make identity, policy checks and event reporting simple enough that teams integrate controls during agent development instead of after deployment.
- Priority
- P0
- Phase
- MVP
- Primary owner
- Developer Experience
- Status
- Blueprint
Surface area
What you integrate against
- REST/JSON APIs for agents, policies, decisions, approvals, events and reports.
- SDKs for Python and TypeScript first; add Java/.NET based on enterprise demand.
- Pre-action policy-check and brokered-action methods.
- Webhook events for violations, approvals, incidents, registry changes and budgets.
- CLI for registration, policy test, gateway setup and diagnostics.
Contract guarantees
- Publish stable versioning and backward-compatibility rules.
- Provide idempotency keys for action/approval APIs.
- Generate typed clients from authoritative schemas where practical.
- Create local developer mode and test tenant fixtures.
pre-action policy check · pythonSDK
from agent_trust import TrustClient
trust = TrustClient(agent_id="support-refunds-01")
# Ask before the agent acts — not after.
decision = trust.check(
action="payment.refund",
resource="billing://customers/8842/refunds",
parameters={"amount_usd": 4200, "reason": "service outage"},
)
if decision.allowed:
result = trust.broker(decision, execute=issue_refund)
elif decision.requires_approval:
trust.request_approval(decision, notify="finance-oncall")
else:
log.warning("blocked: %s", decision.explanation)Definition of done
- A developer can protect a sample agent in under 30 minutes from docs.
- SDK failures return actionable errors and correlation IDs.
- Breaking changes require explicit version migration.
Success metrics
- Time to first API call
- SDK adoption
- API error rate
- Webhook delivery success
DEVELOPER PROMISE
Adding control should feel like adding observability—not rebuilding the agent.