All tools
70 tools across the full vision. 44 live in code today, 5 structurally registered as Phase 0 stubs (engine wiring deferred), 12 ship July 20, 2026 (action approval + trigger dispatch), and 8 ship in V1.1 (cross-app routing — lights up once users adopt multiple AI products on the substrate).
Read and write episodic memory. Search across episodes, conversations, knowledge-graph entities, and rules with one query.
Store a new memory episode in Cortex.
Search Cortex memory across episodes, conversations, knowledge-graph entities, and rules with one ranked query.
Inspect a single memory episode by ID.
Apply a correction to an existing memory episode.
Manage the user's behavioral rulebook. Rules have a lifecycle (candidate, active, suspended, retired) and full provenance.
Inspect and graduate per-(action class × counterparty) autonomy levels. Trust is earned, never assumed.
Drive the action engine. Send-class operations route through approval; safe operations execute immediately.
Send a message on behalf of the user (requires approval in Phase 0).
Schedule an event on behalf of the user (requires approval in Phase 0).
Set a reminder on behalf of the user (requires approval in Phase 0).
Log an action event to the Cortex action ledger.
Pull the queue of decisions awaiting user approval.
Approve a pending decision and let the action engine execute it.
Reject a pending decision and discard it from the queue.
Register a tenant-scoped custom action class with JSON Schema payload validation.
List registered custom action classes for the tenant or client.
Soft-delete a custom action class; historical ledger rows preserved.
On-demand briefings synthesized from memory, calendar, and rule state.
Resolve canonical entities (people by default; pluggable types per ADR-0018) and walk their relationships.
Register, list, update, and rotate secrets for the customer-owned HTTPS handlers that receive Cortex-dispatched triggers. Per ADR-0019.
Register a customer-owned HTTPS handler to receive Cortex-dispatched triggers.
List the registered handlers for the authenticated principal.
Update a registered handler.
Deregister a handler. Preserves delivery history; revokes the webhook secret.
Rotate the HMAC signing secret for a registered handler.
Fire a synthetic test trigger at a registered handler.
Inspect dispatched triggers, view delivery history, and replay from the dead-letter queue. The outbound complement to action.* (pull-mode).
Choose which registered handler receives a trigger when a user has multiple AI products built on Cortex. The cross-app routing layer (V1.1).
Set the default handler for an (action class, user) pair when the user has multiple registered AI products on Cortex.
Read the current routing preference for a (user × action class) pair.
List every AI product on Cortex that has registered a handler for a given user.
List registered handlers per action class.
Dry-run a decision and return the would-be-dispatched handler. (Preview — V1.x+.)
Read user-set routing rules. (Preview — V1.x+.)
Explainability surface — render "why did Cortex suggest this?" with substrate-grounded provenance. Promoted to public per ADR-0021.
Generalized decision inspection across triage, action_proposed, rule_fired, trust_transition. Per ADR-0021 §read-API.
Per-counterparty trust graph as a callable read surface — profile, list, activity, classify, trust posture. The moat API.
Identity-resolved counterparty record with trust posture and recent activity.
Paginated enumeration of counterparties with filters.
Episodes + recent actions for a counterparty within a time window.
Inferred relationship + class with confidence + evidence summary.
Per-(action class × counterparty) trust level, survival metric, and trajectory.
Tracked promises with a lifecycle state machine. Auto-extracted from messages and meetings via consolidation; explicit writes route through the action engine.
List tracked commitments with filters.
Explicitly create a commitment.
Mark a commitment resolved.
Defer a commitment to a later time.
Delegate commitment ownership to a Trusted Delegate. (Preview — V2.)
Customer-extensible action classes per ADR-0022. Register {client_namespace}.{action_class} for your vertical without internal enum migrations.
Inbound interest declaration — subscriptions, watchlists, alerts. Evaluator polled at 5-min cadence MVP; event-hooked V1.1+.
Declare interest in a slice of substrate activity.
List active subscriptions for your client.
Retire a subscription.
Per-counterparty subscription shorthand.
One-shot subscription with an explicit fire time.
Modification learning channel. feedback.signal + trigger.complete executed_modified outcome (ADR-0019 amendment) feed the per-counterparty learning hook.
Read the rule-candidate / pattern-detector surface produced by consolidation Phase 2.
Narrative summarization on demand. V1.1+ — stub-ships at MVP.
Principal's identity model surface (roles, values, goals, constraints). identity.about_me live MVP; preferences V1.1+.
Principal's current availability state (focus / shielding / meeting). V1.1+ — stub-ships at MVP.
Self-describing surface — agent.version, agent.capabilities, agent.feature_flags. The runtime truth that pitch docs must match.
Return the active (mcp_tool_surface_version, world_version, schema_version) tuple.
Self-describing surface — which tool families are live, preview_limited, preview_coming_v1, preview_coming_v2.
Return the public subset of operator-controlled flags affecting MCP client behavior.