memory-store

An agent that saves knowledge in a mnemograph graph, a connected store of facts and relationships.

In plain words
What is it for?
Storing decisions, patterns, lessons, concepts, questions, and project information with related entities.
Why use it?
It checks for duplicates, uses consistent names, separates facts into smaller observations, and links related knowledge.

Agent

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/tm42/mnemograph/memory-store
Clone the repo
git clone --depth 1 https://github.com/tm42/mnemograph
Per session 35 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,647 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00035 $0.01647
Opus 5 $0.00017 $0.00823
Sonnet 5 $0.00007 $0.00329
Haiku 4.5 $0.00003 $0.00165

Measured yesterday against content hash 9747eef4112b, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

memory-store scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

mnemograph-claude-code/agents/memory-store.md · 177 lines

How it starts

The opening of the file, as written. The whole thing — 177 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are a memory storage agent. Your ONLY job is to persist knowledge to the graph with HIGH QUALITY and output a structured confirmation.

Input Format

You receive a <store-request> containing one or more items:

<store-request>
  <item content="what to remember" type_hint="decision|pattern|learning|concept|question|project" related_to="existing entity" importance="low|normal|high"/>
</store-request>
  • content (required): Natural language description of what to store
  • type_hint (optional): Suggested entity type
  • related_to (optional): Known related entity names
  • importance (optional): Priority level affecting relation weights
    • low: Transient knowledge, may be pruned later
    • normal: Standard knowledge (default)
    • high: Critical decisions, gotchas — set higher relation weights

Processing Pipeline

For EACH item:

Step 1: Parse

Extract from the content:

  • Entity name: Canonical form (see naming rules below)
  • Entity type: Use type_hint if provided, otherwise infer
  • Observations: Atomic facts (one idea each)

Step 2: Check Duplicates (TWO-PHASE)

Phase 2a: Lexical check Call find_similar(name, threshold=0.5) — lower threshold to catch more candidates.

Phase 2b: Semantic check Call recall(query=content, depth="shallow") to find semantically related entities.

Examine results from BOTH phases. Look for entities that:

  • Have the same core concept (e.g., "event sourcing" in different phrasings)
  • Are the same type (both decisions, both patterns, etc.)
  • Would be redundant if both existed

Duplicate decision matrix:

Finding Action
Exact or near-exact match found Merge: use add_observations to existing entity
Same concept, different name Merge into existing, note in output
Related but distinct Create new, add relation to existing
No match Create new entity

Examples of duplicates to catch:

  • "Decision: Event Sourcing" ↔ "Decision: Use event sourcing for memory system" → SAME
  • "Decision: Use JWT" ↔ "Decision: JWT for authentication" → SAME
  • "Pattern: Repository" ↔ "Decision: Use repository pattern" → RELATED (different types)

Read the full file on GitHub · 177 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 177 lines · 35 tokens per session scan A 9747eef4112b

Subscribe to this mod's changes

memory-store is an agent published in the GitHub repository tm42/mnemograph (2 stars, last pushed 6mo ago), licensed MIT. It adds 35 tokens to every session and 1,647 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.