Tool Reference

Tool Reference

Complete list of tools available in cortex-engine. As of v1.0.0, all cognitive tools are built into core — 59 tools total, no separate installs needed.

The @fozikio/tools-* npm packages are deprecated. They were absorbed into the engine at v1.0.0 and are no longer maintained — do not install them. See FAQ for details.

Tools are grouped by the engine's own category taxonomy. Only a 12-tool core set (context, query, recall, neighbors, observe, wonder, speculate, feedback, stats, ops_append, ops_query, ops_update) is active by default; the rest are enabled per-namespace via configuration.

memory (11)

Tool	What it does
context	Tiered memory loader: L0 (top-3 names, ~100 tokens), L1 (semantic top-15 + graph edges, ~2k tokens), L2 (multi-anchor full recall)
query	Returns memories ranked by semantic similarity, with HyDE expansion, graph spread activation, and FSRS retrievability weighting
feedback	Records whether a retrieved memory was helpful, adjusting confidence asymmetrically (+0.05 helpful / -0.10 unhelpful)
observe	Records a declarative observation — duplicates merge; high-novelty entries can become memories immediately, others queue for dream
wonder	Records an open question as an interrogative observation, kept separate from factual memories
speculate	Records a hypothesis as a speculative observation, excluded from default query results until validated
recall	Returns recent observations chronologically within a time window, optionally filtered by content_type
neighbors	Returns memories linked to a seed memory via graph edges, with relation type, weight, and evidence
query_cross	Returns memories from sibling namespaces that opt into cross-reads. Read-only
federated_query	Returns memories aggregated from peer cortex instances discovered via the sigil registry
retrieve	Fetches a single memory by id, or runs a plain semantic search if given text

meta (8)

Tool	What it does
stats	Counts and metadata for a namespace — total memories, unprocessed observations, active tools
predict	Returns memories likely to be relevant next, derived from recent observations
reflect	Returns a reflection on a topic grounded in related memories; also stored as a new observation
surface	Returns unresolved cognitive signals — contradictions, tensions, gaps
intention	Prospective-memory reminders that surface when a trigger fires (set / list / fire / cancel)
notice	Stores an observation without embedding for low-latency logging — embedded later in batch
resolve	Marks an open signal resolved with an optional note
query_explain	Query results plus a one-sentence LLM "why" per match. Slower — one LLM call per result

ops (3)

Tool	What it does
ops_append	Appends an operational log entry with type-based TTL (log 90d, instruction/handoff 14d, milestone 180d, decision 365d)
ops_query	Returns operational log entries filtered by project, type, status, or time window
ops_update	Updates an ops entry by id — status, content, or continuity fields (next, blocked)

beliefs (4)

Tool	What it does
validate	Records a prediction outcome and updates FSRS scheduling
believe	Records a belief revision — logs the previous definition with a reason, updates the live memory
belief	Returns the chronological history of a concept's definitions with reasons
contradict	Adjudicates (NLI/LLM) whether an observation contradicts a belief, recording a CONTRADICTION or TENSION signal

consolidation (5)

Tool	What it does
wander	Serendipitous graph walk weighted by information gain — prefers under-explored, uncertain, goal-adjacent nodes
dream	The 7-phase consolidation cycle: cluster, refine, mint, link, FSRS review, cross-domain synthesis, narrative summary. Heavyweight
digest	Ingests a document — extracts facts as observations and generates reflections
ruminate	Free-writes from accumulated context, then optionally extracts beliefs, speculations, and questions
abstract	Proposes a higher-level concept subsuming 2-10 memories. Read-only

threads (4)

Tool	What it does
thread_create	Creates a thought thread for a question or topic spanning multiple sessions
thread_update	Updates a thread — status (open/active/blocked/parked), title/body, session refs, linked memories
thread_resolve	Marks a thread resolved with a final note
threads_list	Returns threads filtered by status (default open), project, kind, or tag

journal (4)

Tool	What it does
journal_write	Writes or updates a daily journal entry keyed by date
journal_read	Returns one entry by date, or a span of the last N days
evolve	Records an identity evolution proposal — a shift in values, preferences, or patterns
evolution_list	Returns evolution proposals by status (proposed, applied, rejected, reverted)

agents (2)

Tool	What it does
agent_invoke	Runs a task with the configured LLM grounded in cortex knowledge, storing findings back as observations
goal_set	Records a desired future state — the gap creates a forward prediction-error signal biasing consolidation

maintenance (5)

Tool	What it does
forget	Reduces salience and increments FSRS lapses so a concept fades from retrieval — not deletion
find_duplicates	Returns near-duplicate memory pairs above a similarity threshold; merge=true auto-merges
sleep_pressure	Unconsolidated observation count, last dream timestamp, hours since last dream. Read-only
consolidation_status	Last dream summary, quality trend across the last 7 dreams, current sleep pressure
retrieval_audit	Retrieval-trace patterns over the last N days — retried tools, misfiring heuristics

social (4)

Tool	What it does
social_read	Returns the current social cognition model — inferred interaction patterns
social_update	Updates the social signal model — energy, mode, engagement notes, scoped by source
social_score	0-100 engagement-potential score for a post, with a per-factor breakdown
social_draft	A 280-char reply tweet drafted from cortex talking points and prior interactions

graph (4)

Tool	What it does
graph_report	Connectivity report — orphans, most/least connected nodes, edge counts, density by category
link	Creates a typed edge (extends, refines, contradicts, tensions-with, questions, supports, exemplifies, caused, related)
suggest_links	Candidate link suggestions — phrases matching known concepts, with similarity scores. Creates nothing
suggest_tags	Suggested tags — names of graph concepts most semantically similar to the input

content (3)

Tool	What it does
content_create	Creates a content piece (idea, blog draft, social post, article) in idea state
content_list	Returns content filtered by state (idea/draft/ready/published/archived) or type
content_update	Updates a piece — state, body, platform versions, published mark

vitals (2)

Tool	What it does
vitals_get	All current vital dimensions (curiosity, connection, confidence, creative_energy, frustration) with baselines
vitals_set	Updates one vital dimension to a value in 0.0-1.0 with an optional note

---

CLI Commands

Available via npx fozikio <command>, not through MCP:

fozikio init <name>    # scaffold a new agent workspace
fozikio serve          # start the MCP server (stdio)
fozikio config         # view or edit configuration
fozikio agent          # manage multi-agent registry
fozikio digest         # process documents through cortex
fozikio health         # cortex health report
fozikio vitals         # behavioral vitals and PE delta
fozikio anomalies      # detect anomalous sessions (Isolation Forest)
fozikio report         # weekly quality report (memory, graph, ops, threads)
fozikio maintain       # data maintenance (fix data issues, re-embed)
fozikio wander         # walk through the memory graph
fozikio migrate        # clone data between two CortexStore backends
fozikio tools          # list cortex tools by category
fozikio nli            # run the bundled NLI service for contradiction adjudication
fozikio help           # show help

Run fozikio help for the current option flags on each command.