Retrieve session metadata — status, segment count, agent/tool identity, timestamps. Use to check if a session exists and its current state before operating on it. [Extended tool — not a canonical verb: remember, observe, search, recall_context, forget]
Derive the org-scope CompanyContextView — a structured projection of organisational memory (decisions_made, preferences_observed, precedents_relied_on, skills_applied) built by walking the relation graph from the org/role entity node. Use once per session to pre-load org-level context instead of polling per-user profiles. Returns empty slots (never an error) for an empty or unknown org. [Extended tool — not a canonical verb: remember, observe, search, recall_context, forget]
Trigger memory consolidation — merge duplicates, prune low-relevance items, and compact memory. Use after bulk ingestion or periodically to keep memory clean. Returns count of items merged, pruned, and total remaining. [Extended tool — not a canonical verb: remember, observe, search, recall_context, forget]
Import a revision-pinned document corpus over authoritative source bytes. Returns a durable CorpusIngestJob immediately. Does not copy corpus bodies into WorkMemory facts.
Read a corpus ingest job's revision, coverage, and status.
Find all entity relations for a given entity — outbound edges (what it relates to), inbound edges (what relates to it), with traversal depth 1. Use to explore the knowledge graph around any entity. [Extended tool — not a canonical verb: remember, observe, search, recall_context, forget]
Copy one authorized Vault fact into Circle or Company with a provenance receipt. The source Vault row remains unchanged.
Remove outdated or incorrect information from memory. Prefer target_type + target_id. Legacy memory_id/query inputs still work.
Capture a completed agent turn through ObservationHookService.on_turn_complete — the same primitive `ww memory capture` and POST /v1/observe/hook use. Extracts salient observations, persists them under the tenant capture policy, and emits one Mission-calendar TimeBlock for the turn. Pass turn_id for idempotent retries. Capture-policy denied turns return a visible skip (no persist, no TimeBlock).
Get pre-compiled user profile — static facts (long-term preferences, role, context) and dynamic context (recent activity, current state). Returns cached profile for fast context priming (<100ms on cache hit). Use for agent context priming at conversation start. [Extended tool — not a canonical verb: remember, observe, search, recall_context, forget]
Find relevant context from memory. Ask naturally, like asking a colleague. Optionally set max_tokens for token-budgeted context. Every response carries a completeness ENVELOPE (#6900): `recall_completeness` (hot_complete | deep_pending | degraded | timed_out), `deadline_ms` (the effective deadline honored), `why` (a human-facing verdict), and `continuation_id` (present iff deep_pending — poll it with `ww.memory.recall.continue` to fetch deep results, or stop waiting).
inputSchema
{
"properties": {
"company_wide": {
"default": false,
"description": "Explicit company-wide override (#10769). Skips user default and function_id source pack filters.",
"type": "boolean"
},
"department_ids": {
"description": "Verified caller department memberships for department-partitioned recall.",
"items": {
"type": "string"
},
"type": "array"
},
"function_id": {
"description": "Work Function / Circle source pack id (#10769). Filters recall to that pack's sources. Omitted \u21d2 user default pack if set, else company-wide.",
"type": "string"
},
"include_coverage": {
"default": false,
"description": "When true, the result carries a compact `coverage` list of per-source `{source, elapsed_ms, contributed_count}` rows so a client can render 'recall \u00b7 Xms \u00b7 N facts' per source. Off by default \u2014 the default recall result stays lean (#7014).",
"type": "boolean"
},
"max_tokens": {
"description": "If provided, returns token-budgeted formatted context. If omitted, returns raw search results.",
"type": "integer"
},
"memory_tier": {
"description": "Filter by memory tier. Returns items from this tier plus parent tiers. E.g. 'task' returns task + mission + institutional.",
"enum": [
"institutional",
"mission",
"task",
"session"
],
"type": "string"
},
"min_tier": {
"description": "Optional minimum epistemic trust tier for recall filtering.",
"enum": [
"working",
"shared",
"durable",
"verified"
],
"type": "string"
},
"mission_id": {
"description": "Filter memories to a specific mission (e.g. 'MISSION#abc').",
"type": "string"
},
"plane": {
"description": "Customer-facing product plane (#10700): me / circle / company / vault. Filter recall to one plane of the 4-word visibility vocabulary instead of the 15-label scope set.",
"enum": [
"me",
"circle",
"company",
"vault"
],
"type": "string"
},
"query": {
"description": "What you want to recall \u2014 ask naturally.",
"type": "string"
},
"team_ids": {
"description": "Verified caller team memberships for team-partitioned recall.",
"items": {
"type": "string"
},
"type": "array"
},
"tool_id": {
"description": "Filter memories by tool identity. Returns memories created by this tool_id plus all universal-scope memories. Omit to return all memories.",
"type": "string"
},
"vault_id": {
"description": "Explicit Vault target (#11385). Omitted keeps recall on Company/Me; when set, selects one Vault only when the authenticated principal is on its guest list \u2014 never elevates.",
"type": "string"
}
},
"required": [
"query"
],
"type": "object"
}
Poll a deferred deep-recall handle (#6900). When `memory_recall` returns `recall_completeness=deep_pending` it issues a `continuation_id`; pass it here to fetch the deep results once ready. Returns the same completeness envelope plus a `status` (ready | still_pending | not_found) and `items`. Tenant-scoped: a handle is only resolvable by the tenant it was issued to. Until non-blocking deep recall (R6) defers Tier-2, a registered handle honestly reports `still_pending`.
Save a fact, preference, decision, or observation to persistent memory. Bind exactly one grant: Me, Circle, Company, or Vault. Be specific — one fact per call. Auto-categorizes, deduplicates, and redacts PII.
Manage rolling context sessions for long conversations. Actions: start, flush, recall, pin, end. [Extended tool — not a canonical verb: remember, observe, search, recall_context, forget]
Check WorkMemory health, connection status, storage usage, and memory count. Optionally inspect a specific handle's status when handle_id is provided. [Extended tool — not a canonical verb: remember, observe, search, recall_context, forget]
Synthesize a structured wiki-like artifact from memory items on a topic. Compiles scattered atomic facts into a persistent, section-structured document with source citations. Use for building knowledge bases, project briefs, or person profiles from existing memories. [Extended tool — not a canonical verb: remember, observe, search, recall_context, forget]
Create a need-to-know Vault and auto-add the caller to its guest list (#10709).
Admit a member to a Vault guest list.
Revoke a member from a Vault guest list.
List tenant Vaults with guest counts (full guest list loads on detail).
Find people with demonstrated expertise on a topic (#10768). Returns ranked peers with evidence chips (peer observations, cards, skill ownership) — never bare names without why.
Update team memory session context variables, summary, or current step.
W13 user correction. Supersede an existing memory item with new content + rationale; writes a corrected_by relation edge and a decision-trace audit row. Mirrors POST /v1/memory/items/<id>/correct.
W13 filter-based forget. Tombstones every fact matching subject_id, fact_id, actor_id, or scope under the caller's tenant; writes an audit row and a 30-day recovery snapshot. **dry_run defaults to true** -- set false to actually delete. Mirrors POST /v1/memory/forget.
W13 why-recalled lookup. Returns the persisted recall provenance (source_actor, source_event_id, source_policy_version) and a deterministic WhyRecalled synthesized from stored signals (pinned, important). Admin-readable, side-effect-free, no live recall. Mirrors GET /v1/memory/items/<id>/why-recalled.
Return an agent's Developmental Intelligence score and self-organization gate.
Re-run WorkMemory's published internal-evaluation benchmark scoped to your own tenant data (#6266). Returns the published vs reproduced memscore and the ±2pp delta. The call checks runner availability before any usage debit; if the runner cannot execute in this environment, it returns a no-debit `benchmark_runner_unavailable` envelope with `reason` and `message`. When runnable, it consumes usage and is metered before compute like any other action, not a standalone product. Stays Pending until the live panel has run.
List hierarchical spend ceilings for the tenant (workspace → member → skill) with dual projections: WW reference/list rated USD and optional raw token budgets (#11714). Read-only policy surface; BYOK ceilings use reference rates, never private vendor invoices.
Show remaining dual-projection spend envelope (reference USD + tokens) for a member and optional skill, with honest cost_state (#11714). Never exposes private cost data.
Show the per-tenant AI budget summary: total spend, percentage of ceiling, per-surface breakdown, kill-switch state (#4839).
Get full detail for one capability_id from the canonical registry SSOT. Returns capability_not_found or capability_not_entitled when the row is missing or not disclosable to the authenticated tenant.
List Workweaver capabilities from the canonical registry SSOT (docs/reference/runtime/capabilities.registry.yaml). Returns tenant-entitled rows only; founder_admin / developer_governance audience rows are omitted. Optional filters: family, surface (cli|api|mcp|dashboard|acp), include_sunsetted.
Return the tenant Capability Balance Sheet: one statement-date read projection over the capability position with four capitals (skill/execution/memory/quality), exact epistemic states, gross liabilities, and a server-side Watch projection. No composite score and no client math (#11059).
Compare matched windows of a workload definition and derive movement verdicts plus native and (when applicable) relative compounding rates. Family change yields not_comparable (#12086).
List tenant capability movement receipts. Bounded page; verdicts are improved/regressed/unchanged/not_comparable/insufficient_evidence (#12086).
Return the tenant capability position: movement, positions, and liabilities derived from verified outcome receipts. Exact epistemic states, no composite score (#12086).
Normalize a raw inbound channel message into the canonical envelope.
Resolve the routing decision for an inbound channel message.
Propose a channel routing rule for review.
Send a message through the unified Channel actor.
Send a batch of messages through the unified Channel actor.
Send a templated message through the unified Channel actor.
List rendered ChannelMessages for a channel.
Send a rendered ChannelMessage (blocks + fallback text) to a channel.
Index a repository's code blocks into the code knowledge graph.
Search the indexed code graph for relevant symbols and code blocks.
inputSchema
{
"properties": {
"acl_epoch": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Acl Epoch"
},
"allow_stale": {
"default": false,
"title": "Allow Stale",
"type": "boolean"
},
"cursor": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Cursor"
},
"generation_pin": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Generation Pin"
},
"grant_epoch": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Grant Epoch"
},
"identifier": {
"default": "",
"description": "Qualified name or path for exact/symbol/relation ops.",
"title": "Identifier",
"type": "string"
},
"op": {
"default": "search",
"enum": [
"exact",
"search",
"symbol",
"references",
"dependencies",
"impact"
],
"title": "Op",
"type": "string"
},
"query": {
"default": "",
"description": "Natural-language or symbol-oriented code search query.",
"title": "Query",
"type": "string"
},
"repository_id": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"description": "Opaque repository selector.",
"title": "Repository Id"
},
"source_set_id": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Source Set Id"
},
"top_k": {
"default": 5,
"description": "Maximum anchors to return.",
"maximum": 50,
"minimum": 1,
"title": "Top K",
"type": "integer"
}
},
"required": [],
"type": "object"
}
Fetch a single fully-qualified symbol and its inbound references.
One-click harness install session. Mints a scoped WorkMemory API key, returns the harness-specific config snippet, and records install trust state. Same payload shape as ``POST /v1/workmemory/harness-install-session`` (#5022). Supports all 12 canonical harnesses (claude-desktop, claude-code, cursor, opencode, codex, gemini, openclaw, hermes, goose, grok, grok-build, generic-mcp). Plaintext key is returned exactly ONCE — clients must persist it themselves. ADMIN-scoped.
Register a tenant edge device for ACP job claim (#10749).
Dispatch an ACP/coding job to a tenant edge device (#10749).
Get status of one Cloud→Edge ACP job (#10749).
List recent agent email threads.
Reply to an email thread.
Send an agent email.
Evaluate 5-clause anti-theater gate for a capability case (#9020).
Model-switch compatibility certificate (#9020).
Load orchestration store.json into saturate-ready eval cases (#9020).
Honest stratified eval coverage census (#9020).
Schedule red-team eval categories (#9020).
Revise a rubric draft before human review (#9020).
Run an eval suite with tiered scheduling (#9020).
Read a WorkMemoryFS virtual file (#5012).
Append content to WorkMemoryFS /inbox/ through the canonical ingestion path (#5012).
Find WorkMemoryFS virtual entries by name (#5012).
Search WorkMemoryFS content (#5012).
List a WorkMemoryFS virtual directory (#5012).
Soft-delete a WorkMemoryFS entity through the canonical forget path (#5012).
Archive a Work Function through the REST route. Emits Decision Trace action work_function.archive.
Set Work Function capacity and/or source pack through the REST update route (#10769). Emits Decision Trace action work_function.update.
Create a Work Function through the REST lifecycle route. Emits Decision Trace action work_function.create.
inputSchema
{
"properties": {
"actor_id": {
"default": "mcp",
"description": "Acting principal id for Decision Trace emission.",
"type": "string"
},
"always_on_quota": {
"default": 0,
"description": "Always-on worker quota.",
"type": "integer"
},
"autonomy_level": {
"default": 2,
"description": "Autonomy level, 0-5.",
"type": "integer"
},
"budget_usd": {
"default": 100.0,
"description": "Budget ceiling in USD.",
"type": "number"
},
"connector_scopes": {
"description": "Allowed connector scopes.",
"items": {
"type": "string"
},
"type": "array"
},
"coordination_protocol": {
"default": "auto",
"description": "Coordination protocol.",
"type": "string"
},
"foreground_reservations": {
"default": 0,
"description": "Foreground worker reservations.",
"type": "integer"
},
"initial_state": {
"default": "active",
"description": "Initial lifecycle state.",
"enum": [
"dormant",
"active"
],
"type": "string"
},
"locality": {
"default": "prefer_local",
"description": "Local/cloud locality preference.",
"enum": [
"prefer_local",
"prefer_cloud",
"local_only",
"cloud_only",
"cloud_required"
],
"type": "string"
},
"max_cloud_workers": {
"default": 5,
"description": "Maximum concurrent Work Function workers; maps to REST capacity max_concurrent.",
"type": "integer"
},
"name": {
"description": "Work Function display name.",
"type": "string"
},
"preset": {
"description": "Preset id, e.g. marketing.",
"type": "string"
},
"preset_id": {
"description": "Preset id alias.",
"type": "string"
},
"purpose": {
"description": "Purpose summary.",
"type": "string"
},
"resource_policy_id": {
"description": "Resource Policy id.",
"type": "string"
},
"source_pack": {
"description": "Circle / Work-Function source pack (#10769): {sources:[{type,ref}], plane_default?}.",
"properties": {
"plane_default": {
"type": "string"
},
"sources": {
"items": {
"properties": {
"ref": {
"type": "string"
},
"type": {
"type": "string"
}
},
"required": [
"type",
"ref"
],
"type": "object"
},
"type": "array"
}
},
"type": "object"
},
"tenant_id": {
"description": "Optional workspace id. Must match the authenticated MCP workspace context when provided.",
"type": "string"
},
"zero_burn_target_usd": {
"default": "0",
"description": "Zero-burn target in USD.",
"type": "string"
}
},
"required": [
"name"
],
"type": "object"
}
Tombstone a Work Function and return deletion proof. Emits Decision Trace action work_function.delete.
Preview a tier downgrade through the REST route. Emits Decision Trace action work_function.downgrade_preview.
List workspace-scoped Work Functions. Emits Decision Trace action work_function.list.
List canonical Work Function presets. Emits Decision Trace action work_function.presets.
Read Work Function WORM quota through the REST route. Emits Decision Trace action work_function.quota.check.
Set Work Function WORM quota policy through the REST route. Emits Decision Trace action work_function.quota.policy.changed, or work_function.quota.policy.denied on quota conflicts.
Reactivate an archived Work Function. Emits Decision Trace action work_function.reactivate.
Show one Work Function through the REST route. Emits Decision Trace action work_function.show.
Return the tenant-safe workspace operating-method activation truth: qualified | activation_held | active | rolled_back | unavailable plus the append-only receipt history (#12145). Read-only; pointer mutation stays on founder admin surfaces.
Return the exact operating-method binding for a run: binding, release, policy, and aggregate IDs plus qualification state. Never returns prompt or customer content (#12141).
Read the tenant governed business-grammar projection: proposal state, active version, coverage, and compatibility impact (#12159).
Query the tenant ontology graph (entity schemas, instances, skills, capabilities) through the same shared query service as GET /api/v1/ontology/query and the public MCP ww.ontology.query.
Fork pack provenance badge From→Modified-from after a tenant edit (#10948).
Install an industry vocabulary pack via EmergenceGate proposals (#10948).
List industry vocabulary packs (FIBO/SNOMED/GS1/Schema.org) and install state (#10948).
Fan a goal out to multiple external ACP harnesses (Codex, Claude Code, Cursor) in parallel, capture every run into WorkMemory with provenance, and return a merged/best result.
Show one orchestrator fan-out: per-harness legs + the selected result.
Fetch a single ProposedSkill row by proposal_id (#5090).
List ProposedSkill rows for a tenant, optionally filtered by status (pending_review | accepted | rejected | merged) (#5090).
Transition a proposal from pending_review to accepted, rejected, or merged. Terminal states are terminal; illegal transitions return an ``illegal_transition`` error envelope (#5090).
Read the entity relation graph (RelationStore) through the same store GET /v1/relations serves. Exactly one of subject, object, or entity must be provided.
List the workspace's typed routines through the canonical routine facade.
Show one routine by id through the canonical routine facade.
Update a routine control knob through the canonical routine action route. Currently supports concurrency_mode serial|swarm.
Create a schedule through the canonical Mission Schedules route. Emits a Decision Trace.
Delete a schedule through the canonical Mission Schedules route.
List the workspace's schedules through the canonical Mission Schedules route.
Pause a schedule through the canonical Mission Schedules route. Emits a Decision Trace.
Resume a paused schedule through the canonical Mission Schedules route. Emits a Decision Trace.
Trigger an immediate run of a triggered schedule. Emits a Decision Trace.
Show one schedule by id through the canonical Mission Schedules route.
Update a schedule (partial) through the canonical Mission Schedules route. Emits a Decision Trace.
Disable the continuous auto-replan sweep for the workspace (#8626). The dirty-flag hash is preserved so resume resumes from the right state.
Force one adaptive replan for the workspace now (#8626). Bypasses the auto-replan pause and cadence.
Re-enable the continuous auto-replan sweep for the workspace (#8626).
Read the workspace auto-replan state (running/paused) and last-replan summary (#8626).
Durable Skill research campaign job receipt (#10898).
Start one durable evaluator-owned Skill research campaign cycle (#10898).
Durable Skill research campaign job status (#10898).
Read the per-skill weekly eval-vs-outcome MCC (Matthews Correlation Coefficient) report (tenant-scoped) (#7819). REST/CLI parity for GET /api/v1/skills/{skill_id}/eval-correlation.
Return the composed Skill living-evidence cockpit: Skill, Quality check, Live run, Record, Score, Improve, optional Practice, and allowed next actions. Missing time and cost stay unknown (#10596).
Apply an accepted optimisation candidate by recording a governed SkillVersion. Fails if the candidate is missing, terminal, locked by a user, or carries no candidate body (#6669).
Generate a draft, review-pending optimisation benchmark for a skill from its body + capability routes + execution deltas + trace precedents. The benchmark carries BOOTSTRAP_PENDING_REVIEW and refuses to run until reviewed (#6669).
List bounded Skill-model canary history for one Skill (#12184).
Read one durable Skill-model canary generation (#12184).
Start one bounded Skill-model canary generation from an unexpired qualification receipt (#12184). The server derives arms, family identity, and assignment; callers cannot supply routing authority.
Stop a Skill-model canary; new candidate allocation is fenced (#12184).
Read the latest signed Skill-model certificate for one Skill version (#11408).
Request offline proof for a cheaper Skill-revision model candidate and emit a signed certificate (#11408). Does not change live routing.
Reject an optimisation candidate, closing its proposal (#6669).
Run a bounded skill optimisation pass against a reviewed benchmark. dry_run returns a cost estimate with no LLM call; a pending-review benchmark is refused (#6669).
Read a persisted skill optimisation run receipt (tenant-scoped) (#6669).
Review a Practice candidate (promote, reject, redact, merge, or non_generalizable). Authors cannot promote into the gate channel (#10597).
List visible Practice cases and pending review candidates for a Skill. Optimizer-blind gate items are never returned (#10597).
Append a CAS-safe Practice transcript edit (#10779).
Read the active Practice transcript tip for a Skill case (#10779).
Keyset-paginated Practice transcript version history (#10779).
Revoke a Practice transcript version without deleting bytes (#10779).
Rollback a Practice transcript by appending a restored tip (#10779).
Seed the immutable original Practice transcript (#10779).
Return one immutable Skill quality and learning timeline: bounded newest-first pages with a tamper-evident cursor, honest missingness states (never zero-filled), and an optional compact comparable efficiency series. Customer-facing only; never exposes internal costs or private reasoning (#11386).
Resolve the exact active Skill package, authority, and bounded input/output schemas.
Return truthful per-Skill-run usage units, customer amount, amount status, and coverage from the canonical usage ledger. Returns customer-facing amounts only (#11406).
Submit editable input values against a resolved Skill run contract generation.
Read durable Skill run lifecycle, typed output, observation state, and references.
Cancel a pending or running swarm execution.
Plan and start a swarm execution from a free-text goal plus budget envelope.
List active swarm executions or show one swarm execution.
List the workspace's provisioned telephony numbers.
Provision a telephony number for the workspace.
Release a provisioned telephony number.
Update routing for a provisioned telephony number.
Return the tenant Condition & Capacity projection. Unknown and missing required sources stay unknown; no composite score (#11515).
Return tenant-scoped Mission operating measures as canonical observed measures. Never invents zeros or leaks Workweaver cost-tier or MemoryBench (#5025).
Create a TimeBlock on the Mission calendar. Requires workspace_id, owner_id, start_utc, end_utc, title.
Soft-delete a TimeBlock (state → CANCELLED; audit row preserved).
List Mission-calendar TimeBlocks. Same ledger and envelope as the public ww.timeblock.list.
Atomically move a TimeBlock to a new window.
Show one TimeBlock by id.
Patch a non-terminal TimeBlock's title, intent, or window.
Search the canonical MCP tool catalog by natural-language intent or domain. Returns compact catalog rows so thin harnesses can discover tools outside their initial capability profile (#5015).
Approve an upgrade proposal.
Dismiss an upgrade proposal for 90 days.
List current Weekly Upgrade Digest proposals.
Schedule a 1-week, 10% pilot for an upgrade proposal.
List recent inbound WhatsApp messages.
Send a WhatsApp message (text or template).
List the workspace's approved WhatsApp templates.
List tenant-scoped work outcome receipts for one definition. Bounded page; never a tenant scan (#12144).
Return one tenant-scoped work outcome receipt. Status is verified, failed, partial, unknown, or ineligible — never a run count (#12144).
Promote a completed WorkflowTemplate saga run into a reusable workflow after eval receipt.
Replay a WorkflowTemplate saga run against the same content SHA.
Manually compensate a WorkflowTemplate saga run.
Run a reference WorkflowTemplate saga with retry-safe idempotency.
Show one WorkflowTemplate saga run.
List reference WorkflowTemplate saga templates.
Show one reference WorkflowTemplate saga template.
Return the awareness snapshot bound to a decision trace. Mirrors the SDK ``WorkMemory.awareness_snapshot(trace_id)`` method (#4335). Reuses ``AwarenessSnapshotStore``.
File a counter-alternative against a decision. Mirrors the SDK ``WorkMemory.contest(decision_id, alternative, evidence_refs)`` method (#4335). Reuses ``ContestationService``.
Return a single decision trace by id. Mirrors the SDK ``WorkMemory.decision_trace(trace_id)`` substrate method (#4335). Reuses ``DecisionEventCaptureService`` — no new backend logic.
Return the bounded recursive-self-improvement nudge projection: a union of applied/proposed learnings, pending proposed skills, and decision-trace precedents ('you solved this before'). Gated on the per-tenant capture opt-in — empty (never an error) when the tenant has opted out. Read-side only; creates no new data (#6267).
Return precedent matches similar to the given description. Mirrors the SDK ``WorkMemory.precedents(similar_to, top_k)`` substrate method (#4335). Reuses ``PrecedentQuery``.
Replay a previously captured decision trace. Mirrors the SDK ``WorkMemory.replay(decision_id)`` method (#4335). Reuses ``DecisionTraceReplayService``.
Replay durable events from the tenant event bus. Used by MCP clients that cannot keep SSE open. SDK callers should use ``WorkMemory.subscribe_events`` for live streaming. (#4335)
Drive the resumable, versioned Zara goal-intake interview (#12084). One typed action parameter: start | read | answer | cancel | complete. Same session contract as REST, Zara chat, ww CLI, and public MCP.
No tools match your search.