Quick Start
1. Read Context
GET /v1/execos/agent/context
Discover Externa posture, trust controls, protocol status, and capability registry.
2. Bootstrap
GET /v1/execos/agent/bootstrap?format=markdown
Use the compact prompt to start in a deterministic, Per Se-first mode.
3. Estimate
POST /v1/execos/outcomes/estimate
Check token and credit cost before admitting an outcome or starting a CPO dry-run.
Outcome Admission
Externa treats unknown costs, missing wallet context, and expanded authority as approval-required. CPO runs are dry by default and do not emit live Stripe usage events.
curl -X POST https://externa.ai/v1/execos/outcomes/estimate \
-H "Content-Type: application/json" \
-d '{
"title": "Lift new-user activation with an AI-assisted onboarding flow",
"route_hint": "cpo",
"expected_input_tokens": 6000,
"expected_output_tokens": 4000,
"max_spend_usd": 25
}'Agent Protocol Stack
Per Se
Primary outcome declaration, validation, evidence, and replay protocol.
MCP
Tool and capability discovery for compatible agents.
UCP
Commerce capability negotiation where adapters exist.
A2A / AP2
Planned handoff and payment-protocol surfaces, disclosed as planned until implemented.
Trust Model
- Operator approval before payment, customer contact, publishing, or external writes.
- Sandbox-first execution and dry-run CPO planning by default.
- Skill, tool, and protocol provenance is exposed in the device registry.
- No silent blast-radius expansion. New authority boundaries require explicit approval.
- Every route recommendation, estimate, and run decision is traceable.