100% FREE · OPEN SOURCE · RUNS ON YOUR MACHINE

Memory that AI Agents Love!

A companion memory agent that lets your agents focus and improve while you keep ownership of everything they learn.

Persistent memory for Claude Code, Cursor, Codex, and 14+ other agents, built on the world's first information-theoretic search engine. 100% free, open source, and runs entirely on your machine - no API keys, no vector database, no backend to babysit.

Get Started on GitHub

89.8%LongMemEval

<90msrecall latency

17+integrations

›_memanto · bash

$pip install memanto

Collecting memanto...
Successfully installed memanto-0.2.0

memanto agent create dev-agent

█

> Agent namespace [dev-agent] created.
[OK] Memory nodes are listening.

remember · recall · answer

# 1. Install the CLI
$ pip install memanto

# 2. One-time setup — pick 2 for On-Prem
$ memanto
  Choose your backend
    1  Moorcheh Cloud    (instant, needs API key)
  > 2  Moorcheh On-Prem  (Docker, no API key)
  ✓ Setup complete — server on http://localhost:8080

# 3. Plug it into your agent
$ memanto connect claude-code

# 4. That's it — your agent now remembers
$ memanto remember "We deploy with Docker on port 8080"
$ memanto recall "how do we deploy?"

Runs in Docker on your machine. Embeddings and answers via local Ollama models — nothing leaves your laptop.

memanto · walkthrough

Instant IngestionConflict ResolutionMulti-AgentSemantic TypesBuilt-in RAGZero-CostFreshnessVerifiable SourcesDeterministic SearchTemporal QueriesInstant IngestionConflict ResolutionMulti-AgentSemantic TypesBuilt-in RAGZero-CostFreshnessVerifiable SourcesDeterministic SearchTemporal Queries

Info-Theoretic ScoringSub-90ms32x Compression17 IntegrationsCross-PlatformNo API KeyLocal EmbeddingsConfidence ScoringDaily SummariesAutonomous CategorizationInfo-Theoretic ScoringSub-90ms32x Compression17 IntegrationsCross-PlatformNo API KeyLocal EmbeddingsConfidence ScoringDaily SummariesAutonomous Categorization

context

Never re-explain your codebase

A new session starts where the last one ended - decisions, conventions, and gotchas already in mind.

token tax

0 LLM tokens per write

Others invoke an LLM on every save. Memanto's write path costs nothing - verify it yourself.

recall

Stored → searchable, instantly

No indexing queue. Memories are findable the moment they're written - in under 90 ms.

setup

One pip install. Nothing else.

No vector database, no schema, no rerankers, no backend service to keep alive.

Origin Story

Six gaps in. Six principles out.

We built Moorcheh.ai first — serverless vector search at scale. Building on it, agents still forgot everything between sessions. So we asked Claude what breaks agent memory. It named six gaps. MEMANTO answers every one.

Representative model reply

My memory exists as a static snapshot injected into context, useful, but fundamentally passive. I can't query it, update it mid-conversation, or distinguish ‘I know this’ from ‘I was told this once.’

The gap

Memory dumped as one blob

Relevant results only

Pulls only what the task needs, never the whole library.

The gap

Stale notes weigh as much as new

Freshness first

New facts always outrank older, outdated ones.

The gap

No idea where a memory came from

Verifiable sources

Every memory knows if you said it or it was inferred from somewhere.

The gap

Everything in one flat pile

Smartly categorized

Sorted into 13 clear types, not one layer.

The gap

Conflicting facts live on forever

Resolves contradictions

Conflicts are caught and reconciled the moment they appear.

The gap

Slow, heavy indexing before recall

Instant ingestion

Usable the second it’s written - recall in under 90ms.

$memanto ui

One local dashboard to run your entire memory agent — no cloud, no setup.

Agents & memories
Conflicts & connections
Your on-prem backend
Migrate from Mem0, Letta & more

Try the live demo →

127.0.0.1:8000/ui

terminal

memanto v0.2.0on-prem · localhost:8080

memory core

Claude CodeCLI

CursorIDE

WindsurfIDE

Codex CLIOpenAI

ClineAgent

GitHub CopilotEditor

API AgentsREST·MCP

VS CodeEditor

Gemini CLIGoogle

Prefers TypeScript

Avoids recursive bug

Uses Supabase

Tab indent, 2 spaces

Tailwind v4

Vitest over Jest

Claude Code

Codex CLI

Cursor

Windsurf

Antigravity

Gemini CLI

Cline

Continue

OpenCode

Goose

Roo Code

GitHub Copilot

Augment Code

$memantoconnectclaude-code

Frameworks & Agent Platforms

Feature	Mem0	Zep	Letta	LangMem	MemantoBest
RememberStore agent memories
RecallSemantic search & retrieval
AnswerMemanto onlyLLM-grounded response from memory
Instant IngestionMemories available instantly after write
Conflict ResolutionAutomated contradiction detection
Semantic Memory Types13 built-in memory categories
Multi-Agent NamespacesIsolated memory per agent
No External API KeyBuilt-in LLM proxy, no setup