REST
Versioned HTTPS shapes
The contract defines JSON responses for records, evidence, decision inputs, and updates.
AgentLayer machine contract
One machine contract for reading the Planetary Model.
REST, MCP, A2A, and the repository SDKs use the same access and response rules. Those rules are verified locally. Hosted connection details require an approved handoff.
MCP
Agent tool access
MCP tools use the same read rules, scopes, and refusal behavior as REST.
A2A
Agent calls
A2A discovery and messages expose selected skills with the same scopes and redaction rules.
SDKs
Typed client helpers
TypeScript and Python helpers exist in the repository. Public registry packages are not published.
The protocol reads first. It returns uncertainty, redaction, and access metadata when state is missing or not available to the caller.
Protocol rules
Agents can read accepted state. They cannot rewrite it.
Read first
Agents request stored context and evidence before a customer workflow decides what to do.
Keep the source trail
Responses carry source references, time, uncertainty, and audit context.
Scoped
Access is governed by approved clients, scopes, entitlements, redaction, and rate limits.
No state mutation
Agents do not write directly into the Planetary Model through the protocol.
Response shape
A response includes the conditions around the result.
The envelope below is sanitized documentation, not a live response. It shows how a client can tell what was read, what supports it, and whether anything was withheld.
Sanitized contract envelope
{
"success": true,
"data": {
"schemaVersion": "crowdalpha.agent.v1",
"asOf": "2026-05-05T18:00:00Z",
"answerType": "grounded_world_context",
"status": "READY",
"entity": {
"entityId": "CROWDALPHA_ENTITY_ID",
"state": "CONSTRAINED"
},
"confidence": 0.77,
"uncertainty": 0.18,
"evidence": [
{
"evidenceId": "observation:CROWDALPHA_EVIDENCE_ID",
"observationId": "CROWDALPHA_EVIDENCE_ID",
"type": "world_observation",
"collectedAt": "2026-05-05T17:52:00Z"
}
],
"provenance": {
"interface": "crowdalpha_agent_interface_v1",
"truthLayer": "world_model",
"authority": "stored_worldstate_record",
"llmGeneratedFactsAllowed": false
},
"interfaceProvenance": {
"interface": "crowdalpha_agent_interface_v1",
"truthLayer": "world_model",
"authority": "stored_worldstate_record",
"llmGeneratedFactsAllowed": false
},
"redaction": {
"applied": false,
"rules": []
},
"entitlements": {
"mode": "agent_client_v1",
"requiredScopes": [
"world.read",
"evidence.read"
],
"effectiveScopes": [
"world.read",
"evidence.read"
],
"missingScopes": []
}
}
}Agent flow
Read, verify, then carry the source trail.
01
Identify the entity, package, or workflow.
02
Ask what stale assumption, missing evidence, or hidden exposure may be costly.
03
Read current Planetary Model state through an approved AgentLayer contract.
04
Inspect uncertainty, redaction, and evidence references.
05
Fetch evidence or decision inputs before citing the result.
06
Carry provenance and audit metadata into the downstream workflow.