Make Your AI Remember Between Sessions

You close the terminal, and your AI assistant forgets the project existed. A real story about a reminder that survived days, many sessions, and a model switch -- and surfaced exactly when it was supposed to.

The problem: your AI assistant forgets context

If you've ever searched "claude code remember between sessions" or "why does my AI assistant forget everything," you already know the shape of the problem. Every session in Claude Code, Cursor, or any other AI coding tool starts from a blank slate. The plan you agreed on yesterday, the thing you promised to do next week, the reason you made that call -- gone the moment the session ends.

So every session starts the same way: you re-explain the project, re-state the plan, re-answer questions you already answered. The tools are stateless by design. Your memory becomes the only persistent layer -- and you're the most fallible part of the system.

The real story: a reminder that outlived every session that touched it

This is our own usage, not a hypothetical. Mid-project, deep in a multi-week push, we told our assistant:

"Remind me on Aug 12 to do the optimization sprint and finalize the plan."

That sentence became a single tool call -- memoir_add_next -- and then we genuinely forgot about it. That's the point of writing things down.

Days later, after many unrelated sessions -- different chats, different tasks, even a different model -- a fresh session opened, and the reminder was just there, in the context that loads at session start. Not a bare "do the thing" either: it carried the date, the task, and the why behind it, because that's what was recorded when it was created.

No human remembered anything. No sticky note, no calendar entry, no scroll through old chat logs. One tool call at the moment the thought occurred, and the memory did the rest.

How it works

memoir is a CLI plus an MCP server. Once connected, your AI tool gets memory abilities it can call mid-conversation. The moment you say "remind me later," the assistant records it:

# During any session -- you say it once

you: remind me on Aug 12 to do the optimization sprint
  and finalize the plan

  # The assistant makes one call
  memoir_add_next("Aug 12: optimization sprint --
  finalize the plan (perf work deferred from the
  July push)")
  Added to Next steps

That next step lives in a plain file you own, synced across your machines. Then, at the start of every later session -- any tool, any model -- the goals and next steps load automatically as context. The future session doesn't have to ask; it already knows:

# Days later, a brand-new session

Next:
- [ ] Aug 12: optimization sprint -- finalize the plan
      (perf work deferred from the July push)

And when you need older context that isn't in the auto-loaded summary, the assistant searches for it:

memoir_recall("optimization sprint plan")
Found 2 memories matching "optimization sprint"

More use cases

• An AI assistant that learns from its mistakes -- one production incident, recorded once, never repeated
• Pick up a multi-week project exactly where you left off -- goals and locked decisions that survive dozens of sessions
• Every decision, with the why attached -- a decision log your AI writes as decisions happen