Reflective Memory for AI

keep provides memory infrastructure for AI agents and applications. Store notes, search by semantic similarity, track intentions with now, and analyze documents into discoverable parts — all through a REST API, Python SDK, CLI, or MCP server.

Unlike simple key-value stores, keep gives agents awareness: versioned notes with structured tags, automatic similarity surfacing, meta-tag relationships, and a reflection practice that improves context quality over time.

Get Started

New to the API? The REST API Quick Start walks you through storing notes, semantic search, file uploads, and cross-source retrieval in about five minutes.

Prefer the CLI? CLI Quick Start gets you from install to first search in under five minutes using uv tool install keep-skill.

Key Concepts

Intentions (now)

A single mutable note that represents your agent's current state, goals, and context. Updated frequently, surfaced automatically on every interaction. This is the core of reflective memory.

Semantic Search (find)

Find notes by meaning, not keywords. Every note is embedded on store; search returns ranked results with similarity scores. Supports tag filters and configurable thresholds.

Structured Tags

Key-value tags organize notes by domain, thread, and facet. System tags (_created, _source, _updated) are managed automatically. Tags enable precise filtering alongside semantic search.

Document Analysis

Upload PDFs, markdown, or long text and analyze them into individually searchable structural parts — sections, themes, and relationships — each with its own embedding.

Meta-Tags

Automatic cross-note relationships: similar items, learnings, and previous versions surface as structured metadata on every retrieval, giving agents longitudinal awareness.

Versioning

Every update creates a new version. Full history is preserved and queryable. Diff across versions to see how context evolved.

Continuations

Stateful multi-step memory interactions. A single continue() loop handles queries, ingestion, and delegated work — with automatic refinement, decision support, and bounded execution. Preview in v0.82.

Integration Paths

• REST API — HTTP endpoints for notes, search, tags, versions, analysis, and file upload. Bearer-token auth, JSON responses.
• Python SDK — Keeper class wrapping all operations. pip install keep-skill or uv tool install keep-skill.
• MCP Server — Model Context Protocol endpoint at /mcp. AI assistants discover all keep commands as tools automatically.
• LangChain / LangGraph — BaseStore, tools, retriever, and middleware. pip install langchain-keep.
• OpenClaw Plugin — Three-layer integration: real-time context injection, automatic memory indexing on compaction, and daily reflection via cron.
• Agent Guide — Best practices for AI agents using keep: when to store, what to surface, how to reflect.