mem + remnant

Your agent loses context between sessions.

Every decision, every fix, every "we already tried that" — gone when the conversation ends. memnant keeps it. And flags when it's wrong.

$ npx memnant
that's it · node 20+ · mit license
scroll

You logged a decision three weeks ago. The code changed since then. Every other tool still serves that decision as current. memnant tells your agent it's stale.

01—03 How it works
01
Run one
command

npx memnant — that's it. First time sets up your project. After that, it compiles everything your agent needs to know: what happened last time, what's still true, what went stale.

npx memnant
02
Open your
agent

memnant is an MCP server. Your agent already knows how to use it. You don't learn anything. You don't configure anything. You just build.

Claude Code · Codex · any MCP client
03
Build

Your agent recalls past decisions before re-deciding. Logs new ones. Flags what went stale. Closes the session on its own. Come back in three weeks — it all compounds.

the agent handles the rest
— Why it's different
●
It
remembers

Decisions, framework fixes, session logs — typed records with semantic search. Your agent checks what was already decided before re-deciding. "Why did we choose Express?" is a one-call answer.

not chat history · not a scratchpad · a structured ledger
●
It knows
when it's wrong

When analytics.ts changes after a decision was logged about it, that record is flagged [stale]. Your agent sees it without you telling it. No other memory tool does this.

staleness detection · the thing that makes this different
The only command you type
terminal
$
Under the hood
Local-first
Single SQLite file. Local embeddings. No API calls. No account. Works offline.
One command
npx memnant. Auto-configures your agent on first run. You never learn the tool.
MCP native
Your agent gets the tools directly. Claude Code and Codex work out of the box.
Open
MIT license. Export everything to markdown or JSON. Your data is never locked in.