@claw-stack/managed-agents-memory

The smart memory brain for Anthropic Managed Agents.

Anthropic's Managed Agents ship with memory_stores — a path-based memory primitive with full-text search and versioning. This package bridges those stores to ai-iq, which adds:

• 🔍 Semantic search (sqlite-vec)
• 🧬 Graph intelligence (entities, relationships, causal edges)
• ⏳ FSRS-6 decay + importance ranking
• 💭 Dream-mode reconsolidation (dedup, normalize)
• 🧾 Provenance (citations, derived-from, reasoning)

Use Anthropic's store as the fast in-session cache, and ai-iq as the durable brain that outlives any single session.

Install

npm install @claw-stack/managed-agents-memory
# Also install the ai-iq CLI (Python):
pip install ai-iq

You need:

• ANTHROPIC_API_KEY with Managed Agents + memory_stores beta access.
• memory-tool on PATH (provided by ai-iq).

Quickstart

import { createSessionWithMemory, syncBack } from '@claw-stack/managed-agents-memory';

const handle = await createSessionWithMemory(agentId, envId, {
  seed: { query: 'kubernetes deployment', limit: 10, semantic: true },
  prompt: 'Check /aiiq/* first. Write learnings to /learnings/.',
});

// ... run the session ...

await syncBack({
  storeId: handle.storeId,
  project: 'my-project',
  runDream: true,
});

How it works

1. Seed — queries ai-iq for the top-N relevant memories and writes them to the Anthropic store at /aiiq/<category>/<id>.md.
2. Session — the agent automatically picks them up via the built-in memory_list / memory_read tools.
3. Sync back — after the session, any memory the agent wrote outside /aiiq/* is captured back into ai-iq with full graph + FSRS treatment.
4. (Optional) Dream — runs memory-tool dream for dedup + consolidation.

Why?

Anthropic's memory stores are path-based with full-text search — great for in-session reference, but no semantic search, no graph, no decay, no provenance. ai-iq has all of that. Together you get the best of both: a fast, managed session cache backed by a rich long-term brain.

License

MIT © Kobus Wentzel