Give your agent a memory in 60 seconds
pip install memanto
memanto # pick "On-Prem" (Docker, no account) or "Cloud" (free key)
memanto connect claude-code # or cursor, codex, copilot, gemini-cli, windsurf… 15 tools
That's the whole setup. No code changes, no wrapper, no vector database to run. Your agent now remembers across sessions, and so does every other agent you connect.
Why developers switch to Memanto
⚡ Writing memory is free and instant
Most memory systems call an LLM on every write to extract entities or build a graph. You pay for those tokens, and you wait for indexing before anything can be recalled. Memanto calls no LLM on write. remember returns, and every agent in your fleet can recall it straight away.
In our temporal benchmark against Mem0's agentic pipeline, Memanto spent 0 extraction tokens to Mem0's 134,690, ingested in 0.1 s to Mem0's 48 minutes, and covered 97.2% of golden answers to Mem0's 69.4%.
🧭 It knows what's true now
Your agent said "we're on Postgres" in March and "we moved to MongoDB" last week. A flat vector store returns both, and your agent has to guess. Memanto supersedes rather than appends: the current belief comes first, the old one is kept with its date, and contradictions between agents are flagged for you to review. Ask what's true now, or what you believed on any past date. Those are two different questions, and Memanto answers both.
🔓 It's yours, all of it
Your memory runs on your machine (Docker + Ollama, no account, nothing leaves). It exports as plain Markdown you can read, diff and commit. It imports from Mem0, Letta, Supermemory, Zep and Hindsight with one command. It's fully open source, with no paid tier holding back the useful half.
A memory agent, not a memory API
Every platform will store your agents' memory. None of them will manage it. Managing it across platforms is against their interest. That is the job of a memory agent.
Claude, Cursor, Bedrock and every vector store will happily keep whatever your agents write. None of them will tell you that two of your agents now believe opposite things about your auth service, that a preference from March is outranking a decision from last week, or that the agent you started this morning is about to redo work another agent finished and reverted.
Memanto does six things on its own judgment. Each one is a shipped feature with a command behind it, not a roadmap item.
| What Memanto does | Command |
|---|
| Extracts | Pulls durable knowledge (decisions, preferences, facts, failures) out of conversations instead of archiving whole transcripts. | memanto remember --from-conversation |
| Consolidates | Merges what every agent learns into one shared memory. Duplicates collapse, and repeated observations raise confidence instead of adding rows. | memanto schedule enable |
| Reconciles | When new knowledge contradicts old, it supersedes and keeps the history. Conflicts between agents are surfaced for review. | memanto conflicts |
| Forgets | Expiry runs on policies you write. Expired memories are labeled, auditable and restorable. | memanto policy apply |
| Briefs | Hands a new or returning agent the slice of memory it needs before it acts. | memanto agent bootstrap |
| Moves | Carries memory across tools and vendors through the Open Knowledge Format, so Claude Code, Cursor and your own stack share one memory instead of five. | memanto memory export --okf |
It works while you sleep. Run memanto schedule enable once. Every night it writes a summary, detects conflicts and sweeps expired memories. You come back to a fleet that knows more than it did yesterday, and you didn't have to sort anything.
Works with everything you already run
38 integrations. Most coding agents connect with a single memanto connect <tool>.
| |
|---|
| Coding agents | Claude Code · Cursor · Codex · GitHub Copilot · VS Code · Gemini CLI · Windsurf · Cline · Continue · Goose · Roo Code · OpenCode · Antigravity · Augment · Kimi Code · Grok Build · Pi |
| Agent frameworks | LangChain · LangGraph · LlamaIndex · CrewAI · AG2 (AutoGen) · Google ADK · Amazon Bedrock AgentCore · Pydantic AI · NVIDIA NeMo Agent Toolkit · OpenAI SDK · Vercel AI SDK · Mastra · VoltAgent · Hermes Agent · OpenClaw · Eve · n8n |
| Protocols & signals | MCP server · REST API · Open Knowledge Format · Langfuse (errors and failed evals become memories) · Vapi (voice agents) |
| SDKs | Python · TypeScript / Node.js · Go |
Want several at once? memanto connect multi sets up all your tools in one interactive pass. → Integration guides
See it work
# Monday: backend-agent learns something
memanto remember "Auth migrated to JWT — session cookies deprecated" --type decision
# Friday: review-agent, which never saw that session, already knows
memanto recall "how does auth work"
memanto answer "why did we drop session cookies?" # grounded answer from memory
# Time travel: what did the fleet believe then, and what has changed since?
memanto recall "deployment policy" --as-of 2026-10-01
memanto recall "deployment policy" --changed-since "last 7 days"
# Housekeeping
memanto conflicts # review contradictions found overnight
memanto ui # open the local dashboard
From Python, the client runs in-process with no server to start:
from memanto import Memanto
memanto = Memanto(agent_id="my-agent") # creates the agent on first use
memanto.remember("Alex prefers oat milk.", type="preference")
memanto.recall("what does Alex drink?")
memanto.answer("Does Alex drink dairy?")
Works on macOS, Linux and Windows.
This is the beginning of the README. Read the rest at the original. Read the full README ↗
