Roles that persist
Create and coordinate roles through the API. Untransferred obligations constrain role closure and merging.
AI execution evidence / Bounded evidence pilots
Durable orchestration. Explicit authority. A portable execution record a reviewer can verify offline.
Preserve the recorded history of an AI undertaking across contexts: authenticated actors, state changes, model and tool activity, approvals and interventions.
Start with a $100 seven-day synthetic evidence check. The first five eligible signed pilots remain $500 each for four weeks. Public naming is optional and may be requested separately after successful technical acceptance.
Built for continuity and control
Create and coordinate roles through the API. Untransferred obligations constrain role closure and merging.
Keep source records and attributed actions. Review proposed changes and export a verifiable record. A summary is not a replacement for raw history.
Authenticated principals, scoped authority, revocation and usage ceilings keep permissions explicit. Your application controls how models contribute.
Your application. Your model.
Protected actions require an enrolled human or dedicated machine principal and current execution permissions. A tenant key alone is not identity. The agentgateway path is qualified for one reference echo tool; calls bypassing CarryFrame are outside its recording boundary.
Read the identity and integration guideAuthorization: Bearer <short-lived token>
X-CarryFrame-Tenant: <authorised tenant UUID>
# Machines additionally use the reviewed
# signed companion and service enrollment.
# Never put credentials in URLs or logs.A durable layer beneath your application
Successor contexts work within persistent roles. Accepted records, obligations and inbox messages belong to the orchestration, not to a chat window.
One undertaking. A continuing record.