Access
Developer access
Stable product access for entities, evidence, decision inputs, domain packages, and subscribed updates.
CrowdAlpha AgentLayer
Planetary Model access for authorized agents.
AgentLayer gives authorized enterprise agents, non-enterprise builders, automations, managed fleets, and customer-owned software governed API, MCP, A2A, and SDK access to structured context packets: current state, evidence, blockers, uncertainty, provenance, decision inputs, and next-action posture.
Agents
Agent tools
Structured tools such as world_get_agent_context for software that needs grounded state, evidence, blockers, and next-action posture instead of brittle page reads.
Builders
Builder access
Eligible access paths for approved enterprise and non-enterprise builders creating agents, workflows, notebooks, and local automation.
Enterprise
System integration
Package access, updates, discovery findings, and decision inputs for customer systems and managed agent fleets that need to react when external conditions change.
Why agents need it
Agents are only as useful as the evidence-linked state they can trust. AgentLayer lets them answer known questions and identify blockers before acting from governed product access.
REST
stable /api/v1 contracts
MCP + A2A
read-first agent interfaces
Discovery
blockers + evidence refs
Controls
scopes, redaction, audit
What agents can do
Read the context packet before taking the next step.
Find costly blind spots
authorized discovery agent
Scan approved Planetary Model and customer-private context for stale assumptions, missing evidence, hidden exposure, operational friction, and action blockers before they become losses.
Build an approved implementation agent
approved builder or operator
Use approved access and evidence-backed context packets so the agent can cite returned CrowdAlpha state before calling tools, updating workflows, or briefing a team.
Check a route or infrastructure state
routing or underwriting agent
Search the corridor, request the grounded context packet, inspect contradictions and evidence, then preserve the decision meaning downstream.
Read cover integrity context
insurance or compliance agent
Query the relevant entity, retrieve history and evidence, then return a review-ready package to a human operator.
Explain the drivers
critical-context agent
Start from a record, inspect the returned driver context, and cite the evidence instead of inventing a free-form explanation.
Monitor subscribed updates
workflow automation
Read entitled updates and request evidence before acting on sensitive movement.
Run a bounded scenario
scenario-planning agent
Run approved scenario analysis, label the result clearly, and keep it separate from observed records.
AgentLayer Protocol
One machine contract across REST, MCP, A2A, and SDKs.
The protocol is the governed contract inside AgentLayer: approved clients read Planetary Model records, evidence bundles, decision inputs, and subscribed updates without writing directly into state.
REST and OpenAPI
Versioned JSON contracts for records, evidence, decision inputs, and updates.
MCP and A2A
Agent tool and agent-to-agent interfaces over the same read-first contracts.
Typed SDKs
TypeScript and Python helpers that preserve envelopes, uncertainty, and evidence refs.
Trust controls
Scopes, entitlements, redaction, and audit receipts stay part of the response.
Honest scope today
What an approved agent can verify today vs. what is still in progress.
We do not want potential customers reading more capability into the platform than is actually wired. Below is the honest read on what an external agent following our beta runbook can prove today, and what is still being built.
Verifiable today
- Bearer-token auth chain end-to-end: malformed bearer → 401; wrong-secret hash → 401; revoked key → 401; wrong-scope key → 403
AGENT_SCOPE_DENIED; inactive entitlement manifest → 403AGENT_ENTITLEMENT_DENIED; cross-tenant manifest → 403AGENT_ENTITLEMENT_TENANT_MISMATCH; happy path → 200 + governed envelope. - Action-ready decision packets:
decision_input_v1shape with SHA-256 fingerprint, posture, evidence refs, blockers, TTL, and a non-negotiablenoActivation: truedoctrine pin. Allowed kinds:trade_lane_exposure_review,carrier_concentration_review,corridor_disruption_watch_review. - Customer-private context contract:
customer_private_context_v1scope-proof booleans (e.g.intersectsWatchlistCarrier,watchlistVesselMatchCount) — never the watchlist contents on the wire, never a proof hash that could be reverse-engineered. - First-party SDKs: TypeScript (
@crowdalpha/agent-sdk) and Python (crowdalpha_agent) both exposeworldQuery(); both are exercised against a real loopback FastAPI process in the beta smoke runbook.
In progress
- Self-serve customer-private context upload(watchlists, portfolios, cases). The read-side contract is live; the upload route, ACL model, and operator UI are in progress and currently handled out-of-band.
- Hosted production environment behind
api.crowdalpha.ai. The Terraform shape is ready; external hosted-AWS proof, hosted DB scale benchmark, hosted AgentLayer replay, and production recurring-feed approval are in progress. - Public domain-package marketplace beyond the Shanghai/Yangshan maritime MVP. Hormuz, Black Sea, Suez, Panama, Bab el-Mandeb, Bosporus/Dardanelles, Malacca, and Red Sea/Gulf of Aden corridor packages are seeded as identity-only — feed activation is in progress, behind contract review.
- Managed-agent capacity. The capacity model is ~100 specialists supervising ~40,000 agents across ~1,600 businesses; that is a capacity claim, not current staffing or customer count.
The beta smoke runbook — docs/runbooks/AGENT_BETA_SMOKE.md — walks an approved operator through the seven probes plus the three modes (--mode stubbed, --mode negative-control, --mode real-pipeline). It is the verifiable evidence behind the “for agents” claim.
AgentLayer turns evidence-linked Planetary Model context and discovery leads into an agent-ready product.
Builders use it to ground agents. Enterprises use it to connect CrowdAlpha context and gap-discovery findings to automations, internal copilots, operational workflows, and production systems.