Zaxy

Register a worker session under a parent mission.

Assign a trust tier to an enrolled agent (steward authority required). Admin-gated when an admin token is configured.

Return full replay-only provenance for every fleet memory: origin actor/session, source + gate citations.

Create a governed fleet memory plane and record the creator as its implicit steward. Off unless fleet_enabled; the fleet plane crosses trust boundaries only through the I4 gate and never grants authority (promotions stay non_authoritative). Admin-gated when an admin token is configured.

Enroll an agent in a fleet at a trust tier (enrollment never escalates trust; steward authority required once the fleet has members). Admin-gated when an admin token is configured.

Propose a cross-boundary fleet promotion (kind: skill, outcome, or rule) through the I4 gate. Routes through the governed FleetManager: the actor's trust tier must permit the visibility scope, the crossing is gate-evaluated and cited, and below-threshold or require_review crossings are held pending steward review. Returns the gated result including review_status and promotion_id. No path makes anything authoritative.

Steward review of a held fleet promotion (accepted activates a pending memory).

Return a replay-backed fleet brief: active and pending promotions, enrolled agents, and trust tiers.

Append a typed event to the agent's persistent memory log. Appended state becomes retrievable through memory_checkout, the front door for reading memory back.

At session start, return compact Zaxy memory guidance and the recommended first checkout call.

Discover Zaxy's memory surface: active capabilities, the ambient usage loop, the active tool profile, and which tools remain callable beyond the listed set. memory_checkout is the front door; call this at session start or whenever tool awareness is unclear.

Read directed causal causes of an entity from graph-backed memory.

Read directed causal effects of an entity from graph-backed memory.

The front door to Zaxy memory: call this first, before substantial work, to checkout current, cited, prompt-ready memory state for a session. Start here; every other memory tool is plumbing or power use, discoverable through memory_capabilities.

Score claim confidence from cited support and conflict evidence.

Run one confidence/metacognition operation: claim (score claim confidence), trajectory (list confidence assessments for a claim), reverification (list re-verification needs), known_unknowns (list known unknowns), or record_known_unknown (record a cited known unknown). Additive umbrella over the single-purpose confidence tools; remaining arguments pass through unchanged. memory_checkout stays the front door for reading memory state.

List append-only confidence assessments for a claim.

Run one consolidation lifecycle operation: candidate (append a cited candidate), propose_from_log (generate candidates from Eventloom segments), status (read review-gated counts), or review (append a review). Additive umbrella over the memory_consolidation_* tools; remaining arguments pass through unchanged. memory_checkout stays the front door for reading memory state.

Append a cited, review-pending consolidation candidate event.

Create non-authoritative consolidation candidates from Eventloom log segments.

Append a consolidation candidate review event without promoting authority.

Read review-gated consolidation candidate status.

Re-ingest a human edit of a memory as a cited, non-authoritative memory.corrected event. The original event (target_seq/target_hash) is never mutated or deleted; the correction cites it and carries the edited content + reason, routed through the I4 update evolution gate (auditable). Retrieval surfaces the correction alongside the retained original; the hash-chain stays intact.

Consult the governed memory-evolution policy before applying an autonomous memory change (consolidate, update, forget, rule_generate, promote). Returns whether the change may auto-apply or must be held for review under the configured autonomy tier, and records a non-authoritative, replayable evolution.gate.evaluated event so the decision is auditable. The default tier is auto_with_rollback (auto-apply above threshold, reversible within the rollback window); stricter tiers (propose_only, require_review) are available.

Explain an outcome from cited causal and checkout memory context.

Export a session's memory as a portable, cited bundle any product can consume. Pull-style: the caller selects what to export via the selector; entries carry sealed Eventloom citations. Returns a signed bundle when the server has an export signing key configured and sign=true, otherwise an unsigned canonical bundle.

After using retrieved context, record whether a memory item was useful, stale, corrected, or reinforced. This closes the loop on context surfaced by memory_checkout, the front door.

Experimental metamemory pre-check: predict whether memory_checkout would likely return something for a query, in roughly a millisecond, from in-memory session state only (no embedding call, no graph query). Returns a non-authoritative verdict (likely | possible | unlikely) with its signal breakdown and raw score. It is a cheap prediction about checkout, never a memory answer or evidence, and its calibration against real checkout outcomes is still being measured — when in doubt, call memory_checkout.

Verified forgetting via cryptographic erasure: permanently forget a forgettable memory by destroying its data-encryption key and appending a cited, non-authoritative memory.forgotten tombstone. The target (target_seq/target_hash) must be a forgettable memory sealed as ciphertext (a __zaxy_cipher cell); the erasure routes through the I4 forget gate (auditable). The on-disk ciphertext and its hash are never touched -- the hash chain stays verifiable -- while the plaintext becomes permanently unrecoverable and readers see [FORGOTTEN].

Batch-ingest typed events from an external producer into the agent's persistent memory log. Records each event's producer via 'actor', preserves caller-supplied causal links (parent_event_id, caused_by, id), dedups re-ingest by 'producer_ref', and projects each event so it is retrievable through memory_checkout. Use this instead of many memory_append calls when a tool or bridge replays its own event stream into Zaxy.

Mark a fact as invalid at a given time (bi-temporal update). History is preserved; the correction surfaces in later memory_checkout (front door) results.

List replay-derived known unknowns for a session.

Agent reports an outcome on recalled memory; reinforces the cited memory and, on failure/partial with a lesson, proposes a governed preventive rule through the evolution gate (cited, reversible, audited). Pass `prior` (the agent's prior confidence the memory would help, 0.0-1.0) to scale reinforcement by prediction error (surprise).

Build a non-authoritative planning packet from applicable procedures.

Append a cited, review-pending belief update proposal without promoting authority.

Query the temporal knowledge graph for relevant context. Power use behind the memory_checkout front door: reach for it when you need targeted hits, temporal filters, or pagination rather than a prompt-ready packet.

Record a cited, open, non-authoritative known unknown.

Replay events from a session starting at a sequence number.

List open unknowns, unresolved conflicts, and low-confidence claims needing re-verification.

Reverse a prior memory evolution with a cited, non-authoritative memory.rolled_back event. The target (target_seq/target_hash) must be a reversible evolution event (a consolidation acceptance, a generated/proposed preventive rule, or a human correction); the reversal cites it, routes through the I4 update gate, and undoes the evolution's effect (a rolled-back consolidation acceptance reverts the candidate to its prior review status; a rolled-back rule or correction stops reaching the prompt). Additive and reversible; nothing is mutated or deleted. Fleet promotions are reversed with the fleet's…

Retrieve similar procedure candidates from Skill Memory and consolidation memory.

Append a typed skill lifecycle event and project it into Skill Memory.

Persist Memory Checkout answer candidates as synthesis artifacts and optional outcome feedback.

Record feedback for one Memory Checkout synthesis ledger row.

Retrieve exact Eventloom source chunks with citations.

Finalize a subagent session and return its handoff bundle.