nex_sprint
Read or update the active sprint. Use action "read" for status overview. To update, provide new content.
nex_improve
Self-improvement analysis + memory consolidation + embedding backfill. Actions: analyze (default), consolidate, backfill-embeddings.
nex_export
Export all StudioMeyer Memory knowledge as structured text or Markdown file. format=markdown generates a human-readable, editable Markdown document with Knowledge Graph, Learnings, Decisions, Skills, Sessions.
nex_context
Load rich session context (auto-context primer). Returns: recent session summaries, active sprint, decisions, learnings, relevant skills, evolution status. Use at session start for seamless continuity.
nex_summarize
PERSIST a session summary (NOT a text summarizer). Saves structured metadata (topics, decisions, learnings, mood) so the next session has continuity. Call at session end alongside nex_session_end.
nex_skill
Track, recall, or find recipe for skills — what tools/patterns work for what tasks. Use "track" to log successful patterns, "recall" to find what works for a domain, "search" to fuzzy-find, "recipe" to get best tool-chain for a task.
nex_evolve
Log a self-improvement. When I change my behavior, learn a better approach, or fix a recurring mistake — track it here. This builds my evolution history.
nex_evolve_status
Analyze my self-improvement history. Shows: total improvements, verified vs unverified, improvements by category, suggestions for areas to focus on.
nex_link
Link two decisions together to build a decision graph. Shows how decisions relate: led_to, replaced, depends_on, contradicts, extends, caused_by.
nex_search
Unified search with temporal decay across ALL knowledge (decisions, learnings, sessions, skills, evolution, entity observations). Uses trigram + full-text + LIKE. Recent memories rank higher. Finds "das Ding mit SSL" even if stored as "certbot renewal". Supports question-conditioned retrieval: episodic queries ("was ist passiert?") auto-filter episodic memories, semantic queries ("wie funktioniert X?") auto-filter semantic facts.
nex_entity_create
Create entities in the knowledge graph WITH their facts in ONE call. Use this to register people, projects, systems, tools etc. BEST PRACTICE: always pass `observations` array — skipping it leaves the entity empty and causes the "0 observations on entity" anti-pattern (facts scattered across learnings instead of on the entity itself). Auto-dedupes: exact name+type match, alias match, suffix-pattern ("X (Y)" merged into existing "X" as alias), trigram similarity warning. Idempotent.
nex_entity_observe
Add observations (facts) to entities. Deduplicates automatically — if a similar observation exists (>50% trigram similarity), it is skipped. Use this to record facts about people, projects, systems, etc.
nex_entity_relate
Create relations (edges) between entities in the knowledge graph. Supports bi-temporal validity (valid_from/valid_to), causal relation types (causes, led_to, prevented, triggered, etc.), and evidence/metadata. Idempotent — duplicate active relations are skipped.
nex_entity_open
Open specific entities with ALL their observations and relations. This is the primary way to load context about known objects. Supports fuzzy name matching. Accepts "name" (string) or "names" (array).
nex_entity_search
Search the knowledge graph by entity names or observation content. Uses trigram fuzzy matching + full-text search (German + English). Finds "das mit SSL" even if stored as "Certbot renewal".
nex_entity_deep_search
Graph-Traversal search — find an entity and follow relations up to N hops deep. Supports causal mode to trace cause→effect chains (only follows causal relation types like "causes", "led_to", "prevented"). Example: "Halim" with depth 2 returns Halim → project → server.
nex_entity_history
Bi-temporal timeline — see ALL changes to an entity over time, including invalidated facts. Shows what changed, when it changed, and what replaced it. Like git log for entity knowledge.
nex_entity_invalidate
Soft-delete an observation — marks it as no longer valid (sets valid_to timestamp) instead of deleting. Preserves history. Optionally creates a replacement observation.
nex_entity_graph
Read the knowledge graph overview: entities with observation counts, relations, and statistics. Supports temporal queries (asOf), causal mode (mode=causal), and focused subgraph via entityId+depth.
nex_entity_delete
Delete entities, observations, or relations from the knowledge graph. Use with care — cascade deletes remove all connected data. Accepts entityNames (fuzzy matched) as alternative to entityIds.
nex_entity_merge
Merge two duplicate entities: moves all observations + relations from source → target. Source name becomes an alias on target, source entity is deleted. Use when nex_entity_create warns about duplicates, or to clean up the knowledge graph.
nex_synthesize
Generate GUIDES from learning clusters (NOT a text summarizer). Groups learnings by category, then creates synthesis documents. Different from nex_summarize (session persistence) and nex_consolidate (dedup/merge). Actions: generate (create/refresh syntheses), list (show all guides), search (find by topic).
nex_learn_link
Connect two learnings that together produce a new insight (A + B = C). Creates emergent knowledge from combinations. Actions: link (connect two learnings), links (get links for a learning), search (find linked insights).
nex_learn_search
Usage-weighted learning search. Combines text relevance + trigram similarity + usage frequency. More-used learnings rank higher. Pass query="" (or omit) to list top-used learnings — combine with verified:false to find the unverified backlog. SIDE EFFECT: by default every call increments usageCount on returned rows and auto-verifies them when usageCount reaches 3 AND confidence >= 0.8 (threshold strictly above nex_learn default 0.7, so auto-verify requires an explicit confidence boost). Pass trackUsage:false for read-only calls (evals, debugging, backlog review).
nex_learn_stats
Learning Intelligence stats: total learnings, syntheses, links, top-used learnings, syntheses by topic.
nex_learn_update
Update an existing learning. Change content, category, tags, confidence, project, source, or verified status. Regenerates search index when content changes. Cannot update archived learnings.
nex_learn_archive
Archive or unarchive learnings (soft-delete, reversible). Archived learnings are hidden from all searches. Supports single (id) or bulk (ids[]) operations. Actions: archive (hide), unarchive (restore), list (show archived).
nex_registry
Central registry for ALL tools, generators, MCP servers, services, and output types. Search instantly instead of browsing files. Default action="stats". Actions: register (add/update entry), search (fuzzy find), list (by category/tags), info (full detail), stats (overview).
nex_delegate
Delegate tasks to agents or manage existing tasks. Actions: create (new task), update (change status/result), list (show tasks), search (find tasks), stats (overview). Agents: aklow-ceo, marketing, ops, support, researcher.
nex_goal
Track business goals with measurable targets. Default action="stats". Actions: create (new goal), measure (add measurement), update (change status), list (show goals), detail (goal with trajectory), stats (overview).
nex_health
Run a system health check. Checks DB connectivity, orphaned sessions, stale learnings, overdue tasks, goals behind schedule, contradictions, decay stats. Returns status (ok/warning/critical), checks, stats, and warnings.
nex_communities
Detect communities (connected clusters) in the Knowledge Graph. Returns entity clusters with optional AI-generated labels. Use to understand how entities are connected. Default action="detect".
nex_contradictions
View and manage contradictions in Knowledge Graph AND potential duplicates/conflicts in Learnings. Default action="stats" (safe overview). Entity actions: list, pending, resolve, stats. Learning actions: scan_learnings (finds SIMILAR learnings via embedding cosine similarity — high similarity means same topic, which may be duplicates, agreements, or subtle conflicts — NOT guaranteed logical contradictions, review pairs manually), list_learnings (show found pairs), resolve_learning (mark as: merged, not_contradicting, superseded, coexist), auto_resolve (auto-archives older learning in pairs with >90% similarity).
nex_decay
Manage confidence decay, importance scores, and lifecycle states. Default action="stats" (safe, read-only). Actions: run (process decay + importance + lifecycle), stats (show health), boost (manually boost confidence), lifecycle (show/process lifecycle state distribution).
nex_agent_report
Write or read agent reports. Agents (CEO, Marketing, Ops) write structured reports after jobs. Memory can query these for cross-agent insights. Default action="read". Actions: write (create report), read (list recent), search (find by query).
nex_deduplicate
Scan the Knowledge Graph for duplicate entities and merge them. Prevents KG bloat. High-confidence duplicates (>85%) are auto-merged, others returned as candidates. Default action="scan".
nex_rerank
Re-score existing search results using Haiku LLM for precision. NOT a standalone search — requires observation IDs from a prior nex_search/nex_entity_search call. Use when top results feel off.
nex_feedback
Apply user feedback to observations. Positive feedback boosts confidence, negative decreases it. Self-improving retrieval.
nex_profile
Generate a synthesized knowledge profile from the KG. Shows top entities, key facts, active topics, knowledge gaps. Can be saved for auto-injection into context.
nex_chronicle
Book Chronicle — track the journey for the book. Auto-captures sessions, milestones, narratives. Query, search, export for book writing. Default action="list".
nex_sm_bridge
Read-only bridge to SM Marketing Agent memory. Access learnings, patterns, decisions, post performance, and strategy from a secondary autonomous agent. Configure via NEX_SM_BRIDGE_DB env var.
nex_reflect
Reflect on accumulated learnings, decisions, and sessions. Gathers data from the last N days, finds patterns and clusters, and returns structured material for reflection. After reflecting, pass insights back via the save_insights parameter to store them as new learnings. Inspired by Hindsight Reflect API.
nex_flush
Flush important context to persistent memory before context compaction. Call this when the context is getting long, before session compression, or when you notice important information that has not been saved yet. Runs all saves through the Learning and Entity Gatekeepers for deduplication — safe to call multiple times.
nex_proactive
Get proactive suggestions based on current context, recent activity, and pending items. Call at session start for a smart briefing. Checks: stale learnings, open decisions, knowledge gaps, pattern clusters, skill issues, session continuity.
nex_consolidate
MERGE duplicate episodic memories into semantic facts (sleep-like consolidation). Different from nex_synthesize (generates guide documents) and nex_deduplicate (entity dedup). Also: causal inference via LLM (causal mode examines 2-hop neighborhoods to infer cause→effect relations).
nex_import
Import AI conversation history from ChatGPT, Claude, Gemini, Copilot, or Perplexity into Memory. Parses exports, extracts entities/learnings/decisions via LLM, and writes to Knowledge Graph. Actions: analyze (preview what would be imported), import (actually write to Memory). Supports auto-detection of platform format.
nex_insights
Auto-analysis of your memory: session patterns, learning trends, knowledge graph health, decision velocity, stale knowledge detection. Pro users get detailed entity rankings and hygiene recommendations.
nex_graph_view
Returns the knowledge graph plus a UI hint so MCP-Apps-capable hosts (Claude.ai, VS Code Copilot, Goose, Postman, MCPJam) render it as an interactive radial view in the conversation. Hosts without UI support get the same data as JSON. Same arguments as nex_entity_graph.
nex_recall_timeline
Returns recent learnings + decisions + sessions as a chronological list, plus a UI hint so MCP-Apps-capable hosts render it as an interactive vertical timeline with type/category filter, search, and click-to-expand. Hosts without UI support get the same data as JSON.
nex_session_replay
Walks through one session as a step-by-step trajectory: start, observations, learnings, decisions, end. Returns the events as JSON plus a UI hint so MCP-Apps-capable hosts render an interactive horizontal replay with click-to-inspect.