How Ava's memory works
Not a chat log — a living, local understanding of who you are and what you have decided, built in five layers and kept sharp over months. And why continuity beats a bigger context window.
Most assistants reset every conversation. Ava does not: she remembers you across sessions, devices and weeks — your project, your preferences, the decisions you made and why. This is the single feature that changes what an agent can be, and it runs entirely on your machine.
Five layers, not one log
A chat history is just a transcript. Ava builds understanding in five stages:
- Extract — pulls the durable facts out of each message, in real time.
- Reflect — deeper analysis at the end of a meaningful session.
- Accumulate — corrections and preferences compound over months.
- Analyse — finds themes, contradictions and connections across it all.
- Consolidate — merges and sharpens, so memory gets clearer over time, never noisier.
Why not just a bigger context window?
Stuffing everything into the prompt is expensive, slow, and forgetful the moment the window fills. Ava keeps a curated, structured memory and pulls in exactly what is relevant — so a decision you made three weeks ago is still in context today, without paying to re-read your entire history every single turn.
It is yours, and it is local
Memory lives on your machine, in your ~/.ava folder, human-readable, with no telemetry. You can read it, edit it, export it. Cloud sync is opt-in and off by default. Your work is never used to train models.
What it feels like
You stop repeating yourself. She remembers you chose one library over another and why; that you like small commits; the shape of the thing you are building. Over weeks she gets more useful, not more generic. That is the compounding advantage of an agent that actually remembers — and it is the hardest thing for anyone to copy.