Recall Meridian vs. other AI memory tools
"AI memory" is a crowded category in 2026. Tools differ on where data lives, what they store, and which agents they serve. This page is a factual comparison — the goal is to help you pick the right tool, including the cases where Recall Meridian is not the right tool.
Quick comparison table
| Tool | Storage | Shape | Tool-agnostic | Open source | Evidence / audit | MCP native |
|---|---|---|---|---|---|---|
| Recall Meridian | Local SQLite | Project | Yes | Apache 2.0 | Draft / trusted, ledgers, gates | Yes |
| Mem0 | Hosted cloud (default) or self-host | User / agent | Yes (SDK) | Apache 2.0 core; managed service paid | No formal gates | Via adapter |
| ChatGPT Memory & Projects | OpenAI cloud | User / conversation | ChatGPT only | Closed | No | No |
| Cursor Rules & Memories | Local project files | Project rules | Cursor only | Closed | No | No |
| recallmcp (joseairosa) | Local files | Note-shaped | Yes (MCP) | MIT | No formal gates | Yes |
| Vector DBs (Chroma, Pinecone) | Local or cloud | Embeddings | Yes | Varies | Not the concern | Via adapter |
Comparison reflects publicly documented behavior as of 2026-05. Tools evolve — check vendor docs for current capabilities.
Frequently asked questions
What is Recall Meridian?
Recall Meridian is a local memory layer for AI agents. It stores project decisions,
milestones, bugs, architecture notes, and handoffs as a queryable local knowledge base,
so coding agents like Claude Code, Cursor, and ChatGPT can pick up where the previous
session left off. It is open source under Apache 2.0 and integrates via the Model
Context Protocol (MCP).
How does Recall Meridian differ from Mem0?
Mem0 is a hosted memory layer focused on conversational and user-preference memory,
with a managed cloud service as the default deployment. Recall Meridian is
local-first (data lives in a SQLite file on your machine), project-shaped
rather than user-shaped, and includes explicit evidence states (draft vs trusted),
promotion gates, and tamper-evident ledgers designed for AI coding work. Recall
Meridian uses a three-layer context model (50-token pointer → 3KB summary →
full JSON on demand) to minimize token usage for long-running projects.
How do I install Recall Meridian?
Recall Meridian installs as a single npm package. Requires Node 18 or later. Works on
Windows, macOS, and Linux.
$ npm install -g @recallmeridian/recall
After install, run recall init in any project directory to bootstrap a
local knowledge base. See the GitHub README
for MCP client configuration (Claude Code, Cursor, ChatGPT).
Is Recall Meridian free?
Yes. Recall Meridian is open source under the Apache 2.0 license. There is no hosted
service, no subscription, and no usage fees — your data lives in a local SQLite
database on your machine. A desktop app is planned (cross-project search, live session
monitoring, KB graph view); the CLI will remain free and open source regardless.
What AI clients work with Recall Meridian?
Recall Meridian works with any client that implements the Model Context Protocol (MCP):
Claude Code, Claude.ai, ChatGPT, Cursor, Codex, and
other MCP-compatible agents. It exposes twelve slash commands —
/recall,
/recall-sync, /recall-kb, /recall-milestone,
/recall-analyze, and others — for session handoff, KB management,
milestone tracking, and AI-driven analysis.