Memanto
A memory agent that curates, retrieves, and hands off durable context for coding agents across sessions.
- Source repo
- moorcheh-ai/memanto
- Stars
- ★ 2.4k
- Last updated
- 6d ago
- License
- MIT
- Primary language
- Python
- FA score
- 50/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Cost
- Free tier plus a paid hosted plan
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A developer using Claude Code or Codex who wants a coding assistant to retain project decisions, preferences, and task context after a context reset.
- Not a fit if
- Teams that need fully offline use without installing Docker
- Users unwilling to use the Moorcheh cloud service or an API key
- Projects needing independently validated, production-proven benchmarks
- Source review
- 50/100 · Major gaps
What does this agent do, and when should you use it?
Memanto is a memory-management tool for other agents, centered on the remember, recall, and answer operations. Its Python CLI manages agents and sessions, stores and retrieves memories, detects conflicts, and can launch a local REST API and Web UI. Memories can be organized into 13 built-in types such as instruction, fact, decision, goal, and preference, with temporal queries, export, sync, and migration commands. The backend can run locally through Docker with Moorcheh and Ollama, use the Moorcheh cloud service, or be hosted in a private VPC according to the README. It connects to coding tools including Claude Code and Codex through memanto connect, and includes a TypeScript/Node.js SDK.
With memanto remember, users can store an individual memory, ingest JSON in batches, or use --from-conversation to extract facts from chat logs; memanto upload places PDF, DOCX, XLSX, JSON, TXT, CSV, or Markdown files into an agent memory namespace. memanto recall returns relevant memories using a query, type filters, and temporal options such as --as-of and --changed-since, while memanto answer generates a RAG answer from retrieved memory. The memanto agent, memanto session, and memanto status command groups manage agents, sessions, and runtime state; memanto conflicts and memanto daily-summary support contradiction handling and daily summaries. memanto memory export and memanto memory sync produce structured Markdown or OKF bundles, and memanto migrate imports memory from Mem0, Letta, Supermemory, or OKF. For programmatic access, memanto serve starts a local session-based REST API, and @moorcheh-ai/memanto starts a local server through uvx and exposes a Memanto client for TypeScript.
- A developer using Claude Code or Codex who wants a coding assistant to retain project decisions, preferences, and task context after a context reset.
- A team operating several development agents that needs separate agent memories, session activation, and handoffs through the memanto agent and session commands.
- An engineering group that wants to turn chat logs, a project MEMORY.md, or uploaded documents into searchable memory.
- A technical lead who needs to inspect decisions as of a particular time or identify contradictory memories.
- A user migrating established memory from Mem0, Letta, or Supermemory and wanting Markdown or OKF exports for inspection and transfer.
- An organization that wants to run its memory backend locally with Docker or in its own VPC instead of using only a cloud workflow.
How do you install or deploy this agent?
For local mode, run pip install memanto, then memanto, and choose “On-Prem.” The flow guides Docker and Ollama setup; the README explicitly requires Docker and says no account or API key is needed. For cloud mode, run pip install memanto and memanto, choose “Cloud,” and paste a free Moorcheh API key from https://console.moorcheh.ai/api-keys. Use memanto config backend to switch between local and cloud backends.
How do you use this agent?
After configuration, store and retrieve a memory with memanto remember "User prefers concise answers" --type preference and memanto recall "user communication style" --type preference. To connect a coding tool, run memanto connect <integration-tool-id>, for example memanto connect claude-code. For a local API or interface, run memanto serve and optionally memanto ui; when the server is running, API documentation is available at http://localhost:8000/docs.
What are this agent's strengths and limitations?
- Defines memory work through remember, recall, and answer, with CLI, REST API, and TypeScript/Node.js SDK entry points.
- Provides 13 built-in memory types, temporal retrieval, conflict detection, and version-related capabilities for long-running project context.
- Offers local Docker/Ollama, cloud, and private-VPC deployment choices, plus Markdown/OKF export and migration from several memory sources.
- The README states that stored memories are searchable immediately without separately provisioning a vector database, embedding pipeline, or reranker.
- The documented full local mode depends on Docker and is guided through Ollama setup; users without that runtime need to choose the cloud backend.
- Cloud mode requires a Moorcheh API key, and Moorcheh is the core retrieval engine, so deployment and service choice involve that provider.
- Although the README lists 20+ integrations, the supplied material gives only the command entry point and some names, not configuration or behavioral details for each integration.
- The README reports LongMemEval and LoCoMo results, but the supplied material does not include reproducible evaluation settings or independent validation details.
How does this agent compare with similar options?
The README positions Memanto as an active memory agent rather than passive memory infrastructure and claims higher LongMemEval and LoCoMo results than Mem0, Zep, and Letta. Its migration command also imports from Mem0, Letta, Supermemory, or OKF; the supplied material does not provide feature-by-feature comparisons.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Memanto This agent | 50 · Major gaps | CLIFreemium | ★ 2.4k | 6d ago | Python | Codex · Claude Code |
| Hindsight | 46 · Major gaps | Self-hosted serviceFreemium | ★ 59k | today | Python | OpenAI API · Claude API |
| Deuz SDK | 69 · Some gaps | Library / SDKFree + model costs | ★ 508 | 20d ago | TypeScript | OpenAI API · Claude API |
| MCP Memory Service | 68 · Some gaps | MCP serverFree | ★ 2.1k | 1d ago | Python | ChatGPT · Codex · Claude Code · Claude.ai |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence: SECURITY.md describes authentication, authorization, rate limiting, input validation, and secure defaults, but lacks code-level implementation details. Dependencies include pyjwt and httpx, but no dependency security audit or vulnerability scanning evidence. External effects: CLI can start local server, connect external services, but permission boundaries not explicit. Rollback: memanto forget and versioning exist, but no rollback mechanism. Source attribution: author info in pyproject.toml, but publisher unverified. Deductions: lack of code-level evidence, least privilege not explicit, user confirmation not explicit, data flow transparency insufficient.
Evidence: README and CLI reference are consistent, but some inconsistencies exist, e.g., README claims 20+ agents, but CLI reference lists only 12. Dependencies listed in pyproject.toml, but no availability guarantee. Failure messages: SECURITY.md and README provide some error handling guidance, but no specific failure message examples. Deductions: dependency availability unverified, failure messages not specific.
Evidence: README describes multiple use scenarios (local, cloud, VPC) and provides integration list. Capability boundaries: CLI reference lists commands, but no explicit limits. Trigger precision: commands and options described clearly. Environment fit: supports macOS, Linux, Windows, with Docker and cloud options. Deductions: capability boundaries not explicit, trigger precision may be affected by incomplete documentation.
Evidence: README provides installation instructions, CLI reference, examples, and documentation links. License is MIT, but copyright year 2026 may be inaccurate. Versioning: pyproject.toml uses hatch-vcs, but no changelog. Maintenance responsibility: author info in pyproject.toml, but no maintenance plan. Deductions: missing changelog, maintenance responsibility unclear.
Evidence: README claims high retrieval accuracy (89.8% LongMemEval), but no reproducible benchmark details. Marginal value: provides multiple features like memory management, migration, integrations. Cost-benefit: free cloud tier and local options, but no cost analysis. Deductions: benchmarks unverified, cost-benefit not quantified.
Evidence: README claims benchmarks and paper, but no reproducible steps. Cross-source corroboration: provides Hugging Face and arXiv links, but no independent verification. Fact-inference separation: README distinguishes facts and inferences, but not explicitly labeled. Deductions: claims lack traceability, cross-source corroboration insufficient.
- Publisher identity unverified; exercise caution.
- Benchmark claims (89.8% LongMemEval) lack reproducible details.
- No dependency security audit evidence.
- Copyright year 2026 may be inaccurate.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
integrations/mcp/v0.1.5 MCP v0.1.5
Release notes excerpt
Release Notes for memanto-mcp v0.1.5 Automatically forwards resolved source attribution across all MCP memory read, write, and answer tools, preserves MCP client attribution when generic writer roles are stored, requires memanto>=0.2.31, and adds a container Dockerfile and llms-install.md guide for MCP registry and marketplace listings. New Features & Improvements - Automatic source Forwarding Across All Memory Tools (memanto_mcp/tools.py, pyproject.toml) …View on GitHub ↗ -
Release notes excerpt
Release Notes for v0.2.31 This release unifies caller attribution around --source / source across the CLI, REST API, and SDKs (Python, TypeScript, Go), fixes memanto connect instruction/skill paths and frontmatter across agents (GitHub Copilot, Goose, Cline, Windsurf, Claude Code), and fixes session-cache and OKF import edge cases from v0.2.30. Breaking Changes - CLI: recall and answer replace --tool with --source (-s) (memanto/cli/commands/memory.py, memanto/cli/connect/templates.py) …View on GitHub ↗ -
integrations/claudecode/v0.1.2 Claude Code v0.1.2
Release notes excerpt
Release Notes for claudecode-memanto v0.1.2 This release updates claudecode-memanto for Memanto's Model Context Protocol (memanto-mcp) workflow, adds --project-dir and --agent-id CLI flags matching memanto connect claude-code, fixes user-scoped global permission placement, and removes the obsolete legacy installer shim. New Features & Improvements - MCP-First Claude Code Integration (pyproject.toml, claudecode_memanto/cli.py) …View on GitHub ↗