Memanto

Name: Memanto
Author: Moorcheh (EdgeAI Innovations)

Moorcheh (EdgeAI Innovations)

A CLI-first memory agent built on three primitives — remember, recall, answer — over 13 typed memory categories with conflict detection and temporal versioning. All retrieval runs through Moorcheh's proprietary information-theoretic search engine, which the CLI provisions either as a local Docker container (fully offline, no account) or as a managed cloud backend. External agents reach it via an MCP server, a local REST API, a TypeScript SDK, or one-command `memanto connect <tool>` wiring for a documented roster of coding agents and frameworks.

Storage: No conventional vector database is exposed — memories are indexed by Moorcheh's engine, running either as a free local Docker container or Moorcheh's managed cloud. Agent and session metadata is cached as local JSON files.
Retrieval: Single-query semantic retrieval through the Moorcheh engine, which the vendor describes as embedding-free with no separate reranking stage; typed filtering across the 13 memory categories; temporal queries via --as-of and --changed-since; and an `answer` mode that adds an LLM-grounded pass over retrieved memories.
Self-host: Self-host: trivial
License: MIT
Pricing: The CLI, MCP server, and SDKs are free and MIT-licensed, and fully local Docker mode is free with no account. Moorcheh's managed cloud has a free Builder tier ($0/mo, 10k vectors), a Production tier ($29/mo plus usage credits, unlimited vectors, 99.9% SLA), and custom Sovereign/VPC pricing. · Free + paid
GitHub stars: 1,693
Last release: 2026-07-23
Last commit: 2026-07-24
First catalogued: 2026-07-25

Strengths

Genuinely one-command local setup — no vector database and no embedding pipeline to operate, with Docker auto-provisioned by the CLI wizard
Broad first-party surface confirmed against live docs: 13 coding-agent pages plus 9 framework-adapter pages, not just a marketing count
Typed 13-category memory model with conflict detection and temporal versioning, so contradictory facts are reconciled rather than stacked
Benchmark claims are backed by an arXiv paper with an ablation study rather than only a blog post
MIT-licensed with documented migration paths in from Mem0, Letta, and Supermemory

Watch out

The retrieval engine doing the real work is proprietary and closed-source — the MIT license covers the CLI and SDK wrapper, so 'no vector DB' means a different, undocumented, vendor-controlled index rather than fully inspectable infrastructure
The '100% free, no API keys' framing applies only to local Docker mode; the MCP quickstart and managed path both require a Moorcheh account, and scaling moves you to a $29/mo-plus tier
Both headline numbers (LongMemEval 89.8%, LoCoMo 87.1%) are self-reported by the same team in their own preprint, with no independent reproduction found
Very young — the GitHub org was created 2026-03-23, roughly four months before this card — with a high fork-to-star ratio typical of a trending but unproven repo
The README's 'Claude Code, Cursor, Codex + 14 others' undercounts imprecisely; the live docs actually show 22 integration pages across two categories

Best for

Developers wanting persistent memory shared across several coding agents with zero backend to manage
Teams wanting typed, versioned memory with conflict detection without standing up a vector or graph store
Offline or air-gapped setups that can run the local Docker engine with no account

How it integrates

MCP server (local-stdio)CLI REST API SDK SDK IDE extension (claude-code, cursor, cline, windsurf, continue, codex, gemini-cli, github-copilot, opencode, goose, roo, antigravity, augment)Framework adapter (langchain, langgraph, llamaindex, crewai, hermes-agents, mastra, vercel-ai-sdk, voltagent, openai-sdk)

Benchmark results

No sourced results yet.

Sources

Last verified 2026-07-25 · updated by discover-frameworks