MemoryAtlas

A Go REST API server is the single source of truth, fronted by thin, stateless per-runtime plugins for OpenClaw, Hermes Agent, Claude Code, OpenCode, Codex, and Dify, plus a documented raw HTTP API for any custom client. Two deployment modes: Direct mode talks straight to a TiDB Serverless database with no server to deploy, while Server mode self-hosts the Go binary in front of TiDB or Postgres and adds multi-agent space management and LLM-based conflict merging. The design principle is explicit — plugins stay thin because storage, search, ingest, and policy all live in the server.

Storage
Pluggable backend selected by config: TiDB with a native VECTOR type (the default and what the hosted API runs), or plain self-hosted PostgreSQL with pgvector using the supplied schema. Ingest can run in a smart mode that uses an LLM to extract discrete facts from raw conversation turns before storage.
Retrieval
Hybrid recall combining vector similarity and full-text keyword search against a unified memories endpoint, with results optionally decorated with the original source-turn excerpts. Multi-space 'Space Chain' routing lets recall traverse an ordered pipeline of memory spaces with a configurable confidence stop-threshold.
Self-host
Self-host: moderate
License
Apache-2.0
Pricing
Self-hosting the OSS server is free (bring your own TiDB or Postgres). Hosted tiers are published: Free ($0, 13,000 add and 1,300 retrieval requests/month), Starter ($9/mo), Pro ($120/mo), and custom Enterprise. The vendor notes billing enforcement is still being turned on. · Freemium
GitHub stars
1,173
Last release
Last commit
2026-07-27
First catalogued
2026-07-25

Strengths

  • The portable-core claim is real and precisely documented — every branded plugin is a thin wrapper over one versioned REST contract that any custom client can call
  • Self-hostable on plain PostgreSQL with pgvector, not locked to TiDB, with a supplied schema file
  • The same API contract serves hosted and self-hosted deployments, so moving to your own infrastructure is a base-URL and credential change rather than a plugin rewrite
  • Transparent published pricing with exact request quotas per tier rather than 'contact us' opacity
  • Rich operational surface for a young project: webhooks, multi-space Space Chains, rate limiting, KMS-backed credential encryption, and an A/B benchmark harness

Watch out

  • No versioned GitHub releases or tags at all — there is no way to pin a version or audit what changed between them, and the static meta.json version does not track reality
  • Despite the portable core, most official plugins default to the hosted api.mem9.ai and auto-provision an API key on first run, so installing a plugin opts you into the SaaS unless you deliberately override the endpoint
  • The vendor states billing is 'coming soon' alongside a live numbered pricing page, so free-tier quota and paid enforcement mechanics are not yet proven
  • Commercially backed by PingCAP, and the hosted value proposition is built to sell TiDB Cloud consumption — a reasonable model, but it explains why TiDB rather than Postgres is the default backend
  • High open-issue count relative to stars; not independently triaged here

Best for

  • Teams already inside the OpenClaw ecosystem wanting managed shared memory with minimal setup
  • Multi-agent orgs needing one memory layer across several coding-agent runtimes without writing a plugin per tool
  • Teams that want a hosted start with a genuine self-host exit ramp that does not require an API rewrite

How it integrates

Benchmark results

No sourced results yet.

Sources

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