One memory layer for every AI you use. Your preferences, facts and goals follow you into ChatGPT, Claude and Gemini — your conventions follow you into Cursor, Claude Code and Codex.
I maintain Memo, a memory layer for AI assistants and coding agents. I write TypeScript daily, expose project rules through /api/mcp, and never deploy before typecheck and tests pass.
Works with the AIs you already use
Most agents re-read the entire transcript every turn. Memo stores the handful of things worth keeping, typed and scored, so every request starts with the right context — and you can see exactly why.
Openly licensed BGE embeddings run on the server, BM25 catches exact terms, then importance and recency re-rank. No embedding API keys, no per-token cost.
Paste a conversation or upload a ChatGPT / Claude export. You get atomic memories — preferences, facts, goals, people, skills — each with a confidence score.
A typed store beats a wall of text. Every memory carries a category, importance, source and timestamps — so your prompts stay small.
Personal memory follows you into ChatGPT, Claude and Gemini as a prompt-ready context pack. Coding agents read the same store over one MCP endpoint.
Search, edit, delete and export everything. Memory should be a database your users can audit — never a black box they have to trust.
Conventions, decisions and gotchas stored per repository. A stable key makes every save an upsert, so rules stay current instead of piling up.
Paste a conversation or upload a ChatGPT / Claude export. Memo keeps the durable preferences, facts, goals and people, and drops the chatter.
Personal memories get a category and importance. Code memories get a repository, a type and an optional key, so conventions update in place.
Coding agents call MCP tools on demand. Chat assistants get a prompt-ready context pack you can paste anywhere in one click.
# 1 — extract memories
curl -X POST https://memex-iota.vercel.app/api/extract \
-H "Content-Type: application/json" \
-d '{"text":"I maintain Memo. We always run typecheck and tests before deploying."}'
# 2 — response
{
"engine": "local",
"memories": [
{
"content": "Maintains Memo, a memory layer for AI assistants",
"category": "fact",
"importance": 0.74
},
{
"content": "Runs typecheck and tests before every deploy",
"category": "convention",
"importance": 0.71
}
]
}Search, filter, edit and delete from one dashboard. The graph view shows what clusters together. The playground answers questions using only what your agent actually remembers.
Simple pricing for this reference build. Start free and upgrade when your agents outgrow the free tier.
For prototypes and side projects that still deserve real recall.
For people who use more than one AI every day and want them to agree.
For workloads that need regional storage, SLAs and private models.
Create an account, then add a memory, connect a coding agent or copy a context pack. Google and GitHub sign-in are supported.