keep

Agent memory that pays attention. The foundation of skillful action.

It includes skill instructions for reflective practice, and a powerful semantic memory system with command-line and MCP interfaces. Fully local, or use API keys for model providers, or cloud-hosted for multi-agent use.

uv tool install keep-skill       # or: pip install keep-skill
keep                              # interactive first-run setup (providers + agent hooks)

# Index content
keep put https://inguz.substack.com/p/keep -t topic=practice
keep put "Rate limit is 100 req/min" -t topic=api

# Index a codebase — recursive, with daemon-driven watch for changes
keep put ./my-project/ -r --watch

# Search by meaning
keep find "what's the rate limit?"

# Track what you're working on
keep now "Debugging auth flow"

# Instructions for reflection
keep prompt reflect

---

What It Does

Store anything — notes, files, URLs — and keep summarizes, embeds, and tags each note. You search by meaning, by keyword, and by graph traversal.

Content goes in as conversation, text, PDF, HTML, Office documents, audio, or images; what comes back is a summary with tags and semantic neighbors. Audio files auto-extract metadata (artist, album, year); image files auto-extract EXIF (camera, date, dimensions). Optional vision/transcription enrichment runs when a media provider is configured, and scanned PDFs and images get background OCR when an OCR provider is available.

Full content is only stored if it's short (~3kB, configurable) - everything else gets summarized. Chunked analysis and summarization produces a focused searchable index that references original files or URLs without filling your disk with duplicated data.

This is more than a vector store: selected tags become edges. A tag becomes an edge when its tagdoc declares an _inverse — for example, .tag/speaker declares _inverse: said, so speaker: Deborah on a note makes keep get Deborah show every note where Deborah spoke. Ordinary tags stay as labels; you opt into navigability per tag key. Bundled edges include speaker, author, references, cites, from/to/cc, and git_commit; you can define your own. When you retrieve any note, keep follows its edges and fires standing queries (.meta/*) — surfacing open commitments, past learnings, related files, commit history. The right things appear at the right time. See docs/EDGE-TAGS.md.

• Summarize, embed, tag — URLs, files, and text are summarized and indexed on ingest
• Contextual feedback — Open commitments and past learnings surface automatically
• Search — Semantic similarity, BM25 full-text, and ranked graph traversal; scope to a folder or project
• Tag organization — Speech acts, status, project, topic, type — structured and queryable
• Deep search — Follow edges and tags from results to discover related notes across the graph
• Edge tags — Turn tags into navigable relationships with automatic inverse links
• Git changelog — Commits indexed as searchable notes with edges to touched files
• Parts — analyze decomposes documents into searchable sections, each with its own embedding and tags
• Strings — Every note is a string of versions; reorganize history by meaning with keep move
• Watches — Daemon-driven directory and file monitoring; re-indexes on change
• Works offline — Local models (MLX, Ollama, etc.), or API providers (Voyage, OpenAI, Gemini, Anthropic, Mistral)

keepnotes.ai — Hosted service. No local setup, no API keys to manage. Same SDK, managed infrastructure.

The Practice

keep is designed as a skill for AI agents — a practice, not just a tool. The skill instructions teach agents to reflect before, during, and after action: check intentions, recognize commitments, capture learnings, notice breakdowns. keep prompt reflect guides a structured reflection (details); keep now tracks current intentions and surfaces what's relevant.

This works because the tool and the skill reinforce each other. The tool stores and retrieves; the skill says when and why. An agent that uses both develops skillful action across sessions — not just recall, but looking before acting, and a deep review of outcomes afterwards.

Why build memory for AI agents? What does "reflective practice" mean here? Read our blog for the back-story →

Integration

Run keep (or keep config --setup) and the interactive wizard offers to install hooks for the coding-tool integrations it detects on your system. The other rows below have their own setup commands.

Tool	How to install
Hermes Agent	curl -sSL https://keepnotes.ai/scripts/install-hermes.sh | bash — installs plugin and runs setup. Fully integrated, with additional skills available.
OpenClaw	openclaw plugins install clawhub:keep (or -l $(keep config openclaw-plugin) from a local checkout). After that, keep config --setup keeps the plugin upgraded. Context engine plugin — full memory assembly, session archival, reflection triggers.
Claude Desktop	keep config --setup, then keep config mcpb to generate and open the .mcpb bundle (details)
Claude Code	/plugin marketplace add https://github.com/keepnotes-ai/keep.git then /plugin install keep@keepnotes-ai
VS Code Copilot	export KEEP_STORE_PATH=... then code --add-mcp "{\"name\":\"keep\",\"command\":\"keep\",\"args\":[\"--store\",\"$KEEP_STORE_PATH\",\"mcp\"]}"
GitHub Copilot CLI	Auto-installed by keep config --setup when ~/.copilot/ is present.
Kiro	export KEEP_STORE_PATH=... then kiro-cli mcp add --name keep --scope global -- keep --store "$KEEP_STORE_PATH" mcp
OpenAI Codex	export KEEP_STORE_PATH=... then codex mcp add keep -- keep --store "$KEEP_STORE_PATH" mcp
LangChain	LangGraph BaseStore, retriever, tools, and middleware
Any MCP client	Stdio server with 3 tools (keep_flow, keep_prompt, keep_help)

After install, just tell your agent: Please read all the keep_help documentation, and then use keep_prompt(name="reflect") to save some notes about what you learn.

Manual MCP setup should treat KEEP_STORE_PATH as the source of truth and bake that value into the launched keep command with --store; many MCP hosts do not inherit arbitrary shell environment variables into stdio server processes.

---

Installation

Python 3.11–3.13. Use uv (recommended) or pip:

uv tool install keep-skill        # or: pip install keep-skill
keep                               # interactive first-run setup

The first run launches an interactive wizard: it picks up any API keys you have in the environment, detects Ollama if it's running locally, lets you pick embedding and summarization providers, and offers to install hooks for any agent tools it finds. Re-run anytime with keep config --setup.

Skip the wizard by setting one of:

export KEEPNOTES_API_KEY=kn_...    # Hosted at https://keepnotes.ai — no local setup needed
export OPENAI_API_KEY=...          # Or GEMINI_API_KEY, OPENROUTER_API_KEY — each does both embeddings + summarization
export VOYAGE_API_KEY=...          # Or MISTRAL_API_KEY — embeddings only; pair with ANTHROPIC_API_KEY etc. for summarization
export GOOGLE_CLOUD_PROJECT=...    # Vertex AI via Workload Identity / ADC

Local option without API keys: install Ollama (auto-detected), or on macOS Apple Silicon uv tool install 'keep-skill[local]' for MLX models.

Or use a local OpenAI-API server (llama-server, vLLM, LM Studio, LocalAI) by setting [embedding] name = "openai" with base_url = "http://localhost:8801/v1" in keep.toml — see docs/KEEP-CONFIG.md.

LangChain/LangGraph integration: pip install keep-skill[langchain] or pip install langchain-keep.

See docs/QUICKSTART.md and docs/KEEP-CONFIG.md for the full provider matrix and configuration options.

---

Quick Start

# Index URLs, files, and notes (store auto-initializes on first use)
keep put https://example.com/api-docs -t topic=api
keep put "Token refresh needs clock sync" -t topic=auth

# Index a codebase — recursive, with auto-watch for changes
keep put ./my-project/ -r --watch
# git: 2 repo(s) queued for changelog ingest

# Search
keep find "authentication flow" --limit 5
keep find "auth" --deep                # Follow edges to discover related notes
keep find "auth" --scope 'file:///Users/me/project/*'  # Scoped to a folder

# Retrieve
keep get file:///path/to/doc.md
keep get ID --history                  # All versions
keep get ID --parts                    # Analyzed sections

# Tags
keep list --tag project=myapp          # Find by tag
keep list 'git://github.com/org/repo@*'  # All git tags/releases

# Current intentions
keep now                               # Show what you're working on
keep now "Fixing login bug"            # Update intentions

Python API

from keep import Keeper

kp = Keeper()

# Index
kp.put(uri="file:///path/to/doc.md", tags={"project": "myapp"})
kp.put("Rate limit is 100 req/min", tags={"topic": "api"})

# Search
results = kp.find("rate limit", limit=5)
for r in results:
    print(f"[{r.score:.2f}] {r.summary}")

# Version history
prev = kp.get_version("doc:1", offset=1)
versions = kp.list_versions("doc:1")

See docs/QUICKSTART.md for configuration and more examples.

---

Documentation

Full docs at docs.keepnotes.ai — or browse locally:

• docs/QUICKSTART.md — Setup, configuration, first steps
• docs/REFERENCE.md — Quick reference index
• docs/KEEP-PUT.md — Indexing: files, directories, URLs, git changelog, watches
• docs/KEEP-FIND.md — Semantic search, deep search, scoped search
• docs/TAGGING.md — Tags, speech acts, project/topic organization
• docs/PROMPTS.md — Prompts for summarization, analysis, and agent workflows
• docs/OPENCLAW-INTEGRATION.md — OpenClaw context engine plugin
• docs/KEEP-MCP.md — MCP server for AI agent integration
• docs/AGENT-GUIDE.md — Working session patterns
• docs/ARCHITECTURE.md — How it works under the hood
• SKILL.md — The reflective practice (for AI agents)

---

License

MIT

---

Contributing

Published on PyPI as keep-skill.

Issues and PRs welcome:

• Provider implementations
• Performance improvements
• Documentation clarity

See CONTRIBUTING.md for guidelines.