neural-memory
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {} |
| resources | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| nmem_rememberA | Store a memory. Auto-detects type, auto-resolves contradicted errors (RESOLVED_BY synapse). Use after completing a task, fixing a bug, or making a decision. Don't use for temporary notes (use ephemeral=true) or project context (use nmem_eternal). |
| nmem_remember_batchA | Store multiple memories at once (max 20). Use when saving 3+ memories together. Partial success — one bad item won't block the rest. |
| nmem_recallA | Query memories via spreading activation. Use when you need past context, decisions, or knowledge. Depth: 0=instant lookup, 1=context (default), 2=cross-time patterns, 3=deep graph. Add tags for precision. Use nmem_context instead for broad recent context. |
| nmem_contextA | Get recent memories as auto-injected context. Use for broad task context. For specific queries use nmem_recall. For project-level context use nmem_recap. |
| nmem_todoA | Quick TODO memory (auto-expires in 30 days). Use nmem_forget to close when done. |
| nmem_autoA | Auto-extract memories from text. 'process'=analyze+save, 'flush'=emergency capture before compaction. Use at session end or when processing large text blocks. |
| nmem_sessionA | Track current session state (task, feature, progress). Single-session only. For cross-session persistence use nmem_eternal. |
| nmem_eternalA | SAVE project context, decisions, instructions that persist across sessions. Pair with nmem_recap to LOAD. Use for project-level facts, not task-specific memories. |
| nmem_recapA | 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. |
| nmem_situationA | 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. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| NeuralMemory System Prompt | Instructions for AI on when/how to use NeuralMemory |
| NeuralMemory Compact Prompt | Short version of system prompt for limited context |
TDQS
Scored across 10 tools
The retrieval tools (nmem_situation, nmem_recap, nmem_recall, nmem_context) and the storage tools (nmem_remember, nmem_remember_batch, nmem_todo, nmem_auto, nmem_eternal) heavily overlap, and the descriptions spend significant effort cross-referencing each other ('use nmem_context instead', 'for specific queries use nmem_recall') to compensate. The guidance is helpful but signals genuine boundary blur, so an agent can still misselect among the four recall-style tools.
All names share a uniform nmem_ prefix with lowercase snake_case, which is highly predictable. The only minor deviation is mixing action verbs (remember, recall) with noun-style names (situation, context, eternal, session), but the conventions themselves are consistent.
Ten tools is a reasonable scope for a memory server with distinct storage and retrieval concerns. It is slightly on the heavy side given the functional overlap between recall/situation/context/recap, but each still earns a roughly distinct place.
nmem_todo's description instructs the agent to 'use nmem_forget to close when done,' but no nmem_forget tool exists, creating a real dead end. There is also no delete or update/contradict-resolution tool exposed despite the surface emphasizing memory lifecycle management.