Hey everyone, Majid here, co-founder and CTO.
Memanto is a memory agent for your AI agents it remembers, recalls, and answers, so Claude Code, Cursor, Codex (and ~14 other agents) stop forgetting your project every time the context resets.
The big change we're launching today: the whole stack now runs 100% locally and free. pip install memanto, one command, and it sets up everything on your machine retrieval engine, embedding models, even a local LLM via Ollama. No API key, no vector DB to manage, nothing leaves your laptop. If you prefer zero-install, there's a free cloud option too, and you can switch between them.
We also built one-command migration, so if you're on Mem0, Zep, or Letta, you can bring your existing memories over and compare side by side.
The conflict flagging is the part that stood out to me β most memory tools just append and let the last write win, so having the system surface it when two agents believe opposite things feels like the actual hard problem. Curious how the pre-action brief works in practice with Claude Code: does it inject a short summary at session start, or does the agent pull it on demand? The expiry policy per memory type also sounds right β a preference and a fact shouldn't age the same way.
@jcbuildsΒ Thanks, glad the conflict part stood out.
With Claude Code it's a bit of both. At session start, the important stuff (preferences, key decisions, anything flagged) gets loaded so Claude starts with the right context. Everything else stays in memory, and Claude pulls it when the task needs it. That keeps the starting context small and nothing gets lost.
And yes, a preference like "use tabs" should stick around way longer than a fact like "the build is broken right now," so each memory type ages at its own pace.
Flagging the conflict when two agents believe opposite things, instead of letting the last write win, is a really good call. Stale or contradictory memory is how agents end up confidently wrong.
How do you decide a memory is stale? Is it purely time-based, or can new evidence retire an old belief on its own?
@ssindhiaΒ These are two important questions, but we treat them in two categories.
1. To manage the staleness of memory, Memanto has a detailed memory expiry policy where you can set your rules with high granularity for every type of memory. Then the expiry policy works automatically. You can also expire or revive a specific memory manually if you find it necessary.
2. The conflict detection works across the entire month (or more) of memories and they are flagged automatically on a scheduled basis and brought to your attention. Then you can decide what to do with them. Keep old, keep new, remove both, etc. It is handled case by case to allow for practical situations to be resolved.