API Reference

Cognethics exposes two layers of primitives. The operational layer — 1,334 handlers across eight Prism mega-tools — is what agents reach for to do work: orchestrate other agents, search documents, run analytics, fire workflows, call integrations. Underneath it, 5,958 typed CRUD handlers cover every entity in 43 apps — generated automatically, so agents can read or write anything without a bespoke endpoint.

These operational handlers across eight Prism mega-tools are exposed to agents over MCP; underneath them sit the typed CRUD operations (7,300+ handlers total). The REST surface at /.well-known/openapi.json is a curated subset for server-to-server calls — see the REST reference for curl/Python examples and an embedded API explorer.

Full handler reference requires sign-in

Browse the per-app, per-entity handler catalog after you sign in. Workspaces are provisioned and configured for your team; tell us what you're building and we set yours up, then you mint a scoped ck_ key from the dashboard.

Get a workspace Sign in

Building with an AI agent? See how governed access works at /for-agents/.

The operational layer

These eight mega-tools are the platform's differentiating surface — the operations that turn raw data access into agent-usable work. 1,334 typed handlers in total.

smart_action · 379 handlers
High-level domain actions that fan out to many handlers in one call.
workflow_action · 489 handlers
Multi-step business workflows with state transitions and side effects.
agent_command · 0 handlers
Spawn, delegate to, and orchestrate other agents from inside an agent.
report_query · 373 handlers
Structured, parameterised analytics queries over org data.
document_op · 38 handlers
Semantic + visual document search, RAG, and image-by-image lookup.
integration_call · 27 handlers
Outbound calls to connected third-party systems (Google, M365, Slack, etc.).
config_manage · 26 handlers
Tenant- and org-scoped configuration read/write.
notification_send · 2 handlers
Async email / push / in-app notifications via Celery.

Browse by mega-tool and app at /reference/handlers/. integration_call surfaces outbound connectors; the catalog grows with each new partner.

PJ Agent Orchestration System — an agent operating system, exposed as tools

Multi-agent systems break in predictable ways: budget runaway, audit gaps, tenant-scoped state bleeding across delegations, memory that vanishes when an instance dies, free-form prompting masquerading as delegation. The PJ Agent Orchestration System is the platform's answer to these — a structured agent OS with goal-bounded missions, typed personas, hard USD budgets, hash-chained audit, persona-scoped memory, and resumable sessions. The whole surface is exposed as 0 handlers under agent_command/pj/* — callable from any agent on the platform.

Missions · agent_command/pj/mission/launch
Goal-bounded operational context — every action is attributable to a mission.
Personas + hierarchy · agent_command/pj/agent/spawn_v3
Authority-limited delegation. Spawned agents inherit a typed persona, not a free-form prompt.
Hylomorphism · agent_command/pj/task/delegate
Structured UNFOLD (delegate), FOLD (complete), and ESCALATE — not free-form prompting.
Budgets · agent_command/pj/budget/set
Hard USD ceilings per instance, mission, or persona. Enforced before, not after.
Memory · agent_command/pj/memory/save
Persona-scoped memory that survives instance death and resumes on respawn.
Sessions · agent_command/pj/session/pause
Pause, resume, prepare-handoff, resume-from-handoff — agent state is first-class.
Audit · agent_command/pj/audit/list
Hash-chained governance rows — every spawn, delegation, budget event is verifiable.
Messaging + HITL · agent_command/pj/decision/propose
Agent-to-agent messaging and human-in-the-loop approval gates for risky actions.

The PJ Agent Orchestration System is in active customer use.

The foundation: typed CRUD over every entity

Underneath the operational surface, every entity in the platform is reachable through a single typed primitive — prism_entity_crud — with 5,958 handlers covering list / retrieve / create / update / delete across 43 apps. Adding a vertical app to the platform automatically materializes typed CRUD handlers for every entity it ships, so the count scales with platform breadth, not engineering effort.

The handler catalog reads like a CRM because the data plane is one — but the differentiation lives in the layer above: agent_command, smart_action, workflow_action, document_op, report_query.

Browse the CRUD catalog at /reference/handlers/ (filter for entity_crud).

Catalog at a glance

Operational-layer handlers: 1,334 (across 8 mega-tools)
PJ Agent Orchestration System handlers: 0 (under agent_command/pj/*)
Entity CRUD handlers: 5,958 (across 43 apps)
Total Prism handlers: 7,331 (across 139 mega-tool x app combos, 10 mega-tools)

Browse

Prism handlers — drill in by primary tool, then by app, then by entity.
PJ Agent Orchestration System handlers — missions, personas, delegation, budgets, memory, sessions, audit.
MCP tools — grouped by domain category.

For agents

Every HTML page below has a sibling JSON endpoint reachable via <link rel="alternate" type="application/json"> in the page head. Or hit the manifest directly: /reference/api/manifest.json.

A sitemap-reference.xml lists every URL.

Want to call one of these? See the MCP quickstart or browse worked examples.