Biological memory for AI

Give your agents persistent context that naturally learns, associates, and forgets.

Conch is an embedded memory engine that learns, associates, and forgets over time. No cloud. No API keys. Just a single, highly intelligent SQLite file.

$ cargo install conch

Scale intelligence infinitely.

Context scalability

Provide your agents with years of history without polluting the prompt. Token utilization stays perfectly efficient, no matter how much your agent learns.

Semantic precision

Locate exact meaning across all stored memories instantly. Completely resilient to variations in human phrasing or formatting.

Dynamic relevance tracking

Give your memory a sense of temporal weight. Critical user preferences endure indefinitely, while obsolete logs and episodes organically fade away.

Knowledge consolidation

Stop generating duplicated messes. Conch automatically reinforces frequently accessed memories while avoiding endless duplication.

It learns what matters. And forgets what doesn't.

Just like a human brain, Conch strengthens memories dynamically as they are used. Irrelevant episodes fade quickly, essential facts endure, and useless noise is garbage-collected forever.

Jared works at Microsoftauth: 0.94

Recalled 3 days ago — reinforced

Prefers dark modeauth: 0.31

Recalled 40 days ago — fading slowly

Had lunch at Chipotleauth: 0.01

Obsolete episode — pruned permanently

Retrieve meaning. Not just keywords.

Powered by a hybrid BM25 and vector recall engine, fused via Reciprocal Rank Fusion. Your agent immediately finds the exact intent behind a query, completely resilient to phrasing differences.

# Stored initially:
$ conch remember "Jared" "is employed at" "Microsoft"
# Intuitively recalled:
$ conch recall "where does Jared work?"
→ [fact] Jared is employed at Microsoft
score: 0.847 | strength: 0.94

Spreading activation mapping.

Memories don't exist in a vacuum. Retrieving one memory naturally surfaces deeply adjacent knowledge through shared subjects and objects. You are building an intelligent graph of associations.

Jared→ works at →Microsoft
Microsoft→ builds →Copilot
Jared→ works on →Copilot

Just ask OpenClaw to use it. That's it.

Conch operates as a native agent skill. Give OpenClaw the skill URL and define your Mandatory Storage Triggers in AGENTS.md, and your agent will deterministically learn your project context.

# 1. Provide the skill instructions:
User: Read https://raw.githubusercontent.com/jlgrimes/conch/master/skill/SKILL.md and install conch.
# 2. Setup Deterministic Storage (AGENTS.md):
### Mandatory Conch TriggersYou are a system, not a person. These are IF-THEN rules, not aspirations.| Condition | What to store ||-----------|--------------|| A project is named | name, concept, stack, repo, location || A tech decision is made | what was decided and why || Something is built and pushed | repo URL, local path, status || A preference is expressed | the preference, verbatim |
Open Source & MIT

Evolve your agents.

$ cargo install conch