Neural Memory

SAVE project context, decisions, instructions that persist across sessions. Pair with nmem_recap to LOAD. Use for project-level facts, not task-specific memories.

Add evidence for/against a hypothesis. Bayesian confidence update with auto-resolve. Requires an existing hypothesis_id from nmem_hypothesize.

Long-term brain growth trends: maturation, plasticity, coherence. Use for trend analysis, not immediate health (use nmem_health for that).

Explain how two concepts connect in the neural graph (shortest path with synapse types and weights). Use to understand relationships between entities.

Delete a memory. Soft delete by default; hard=true for permanent removal. Use to close completed TODOs or remove outdated memories.

Track what the brain doesn't know. Detect gaps from contradictions, low-confidence, or recall misses. Resolve when new info fills them.

Goal-directed recall — manage active goals that bias memory retrieval toward relevant context. Active goals make recall relevance-based (proximity to goal) instead of just similarity-based.

Learned workflow habits from tool usage patterns. Suggest next action, list habits, or clear.

Get brain health diagnostics including grade and recommendations.

Create or inspect hypotheses (Bayesian confidence). Cognitive workflow: hypothesize -> evidence -> predict -> verify -> cognitive (dashboard). Auto-resolves at >=0.9 (confirmed) or <=0.1 (refuted) with 3+ evidence.

Import memories from external systems (ChromaDB, Mem0, Cognee, Graphiti, LlamaIndex). One-time migration tool.

Index codebase for code-aware recall. Extracts symbols, imports, and relationships. Run once per project, re-scan after major changes.

Retrieve full content of a previously offloaded tool result by its ref_id (returned from nmem_offload). Returns the original raw content. Returns an error if the ref has expired or never existed.

Manage memory lifecycle: view compression states, freeze/thaw individual memories, recover compressed content. Use when a memory was incorrectly compressed.

Brain growth milestones (100, 250, 500...10K neurons). Check for new achievements, view progress to next, or generate growth report.

Generate memory narratives: timeline (date range), topic (spreading activation), or causal chain. Use to understand how knowledge connects.

Store a large tool result as an ephemeral neuron (24h TTL) and return a compact summary + ref_id. Use when tool output is large (>2KB) and you may need to inspect it again later without keeping it in context. Drill back into full content via nmem_inflate(ref_id).

Pin memories as permanent KB (skip decay/pruning/compression). Use for critical knowledge.

Create falsifiable predictions linked to hypotheses. Use nmem_verify to record outcomes (propagates evidence back to hypothesis).

Pro: Smart merge — HNSW-accelerated memory consolidation. Finds clusters of semantically similar memories and merges them, keeping the highest-priority memory as anchor. Use dry_run=true to preview merge plan without executing.

Trace or audit a memory's origin chain. Use when verifying where a fact came from or adding verification/approval stamps.

Query memories from NeuralMemory. Use this to recall past information, decisions, patterns, or context relevant to the current task.

LOAD project context saved by nmem_eternal. Call at SESSION START to restore cross-session state. Level 1=quick (~500 tokens), 2=detailed, 3=full.

Refine an instruction/workflow: update content, add failure modes, add trigger patterns. Use to improve instructions based on real-world execution outcomes.

Pin/unpin neurons as reflexes (always-on in every recall). Reflexes bypass spreading activation and appear first in context. Use for critical rules, preferences, or constraints that must always be present.

Store a memory in NeuralMemory. Use this to remember facts, decisions, insights, todos, errors, and other information that should persist across sessions.

Store multiple memories at once (max 20). Use when saving 3+ memories together. Partial success — one bad item won't block the rest.

Report instruction execution outcome (success/fail). Builds track record — high success rate boosts recall priority. Call after executing an instruction.

Spaced repetition reviews (Leitner 5-box system). Queue due reviews, mark success/fail, view stats.

Evolve a hypothesis into a new version (SUPERSEDES chain). Use when understanding changes — preserves belief evolution history.

Track current session state (task, feature, progress). Single-session only. For cross-session persistence use nmem_eternal.

Get full verbatim content + metadata + synapses for a memory by ID. Use after recall when you need exact content, not the summarized version.

One-shot snapshot of the current working situation: active session task, top 3 recent decisions, open blockers, gap detection. Replaces nmem_recap + multiple nmem_recall calls when resuming a session. Pure read — never mutates state.

Register external sources (docs, laws, APIs) for provenance tracking. Use before nmem_train to link trained memories to their origin.

Get brain statistics including memory counts and freshness.

Brain Store — browse, preview, import, export, and delete knowledge brains. Share curated brains with the community or import others' expertise.

Autocomplete from brain neurons. No prefix = idle/neglected neurons needing reinforcement.

Knowledge Surface (.nm file) — compact graph (~1000 tokens) loaded every session. Generate to rebuild, show to inspect.

Manual sync with cloud hub. Push local changes, pull remote, or full bidirectional sync.

Configure sync: setup (onboarding), activate (license key), get/set settings.

View sync status: pending changes, connected devices, last sync time.

Backup brain database to Telegram. Requires Telegram bot config.

Auto-tier: promote/demote memories between HOT/WARM/COLD by access patterns (Pro). Evaluate (dry-run) before apply. Free users: manual tiers only.

Pro: View memory tier distribution and compression stats. Shows how memories are distributed across tiers (ACTIVE/WARM/COOL/FROZEN/CRYSTAL) and estimated storage savings.

Quick TODO memory (auto-expires in 30 days). Use nmem_forget to close when done.

Agent tool usage analytics: frequency, success rates, daily trends. Use to understand which NM tools are being used and how effectively.

Train brain from docs (PDF, DOCX, PPTX, HTML, JSON, XLSX, CSV). Pinned by default as permanent KB. Requires: pip install neural-memory[extract].

Train brain from database schema (tables, columns, relationships). SQLite supported.

Copy memories from another brain by tags/types. Use for sharing knowledge between project brains.

Record prediction outcome (correct/wrong). Propagates evidence to linked hypothesis. Use after observing the predicted event.

Brain version control: snapshot current state, rollback, or diff between versions. Use before risky consolidation or major changes.

Generate charts from memory data (Vega-Lite/markdown/ASCII). Use for financial metrics, trends, or any structured data in memories.

Watch directories for file changes, auto-ingest into memory. Scan for one-shot, start/stop for continuous monitoring.