Capability · Memory
Long-term memory that never resets
A private, persistent memory and knowledge graph — tenant-isolated, stored in Zurich, and never used to train shared models.
Most AI tools forget everything when the session ends. monopea keeps a private, long-term memory of your goals, decisions, and working preferences — plus a knowledge graph built from your documents and conversations — so context compounds instead of evaporating.
Persistent memory is also a custody question, which is why it is engineered like one: your memory is tenant-isolated at the database layer, stored at rest in Zurich, inspectable and deletable by you, and never used to train shared models.
Context that carries across every session
Objectives you set, preferences you state, and corrections you make are written to durable memory tied to your workspace, not to a single chat thread. Alongside it, a knowledge graph links the entities and facts from your documents and conversations, so recall is structured — the agent retrieves what a task needs, not a transcript dump.
Pick up tomorrow, next week, or next quarter and the agent already knows the shape of your work: what you decided, what you rejected, and why. The longer it runs, the less you re-brief.
- Goals, decisions, and working preferences persist by default
- A knowledge graph links facts from your documents and conversations
- Recall is scoped to the task — relevant context, not noise
Isolated, inspectable, and never trained on
Your memory is private to your tenant and enforced as such in the database: two layers of isolation — application-level scoping plus Postgres row-level security under runtime-minted tenant tokens. It is stored at rest in Zurich, Switzerland, is never used to train shared models, and you can inspect or delete what the agent retains at any time.
That custody is what makes it safe to let the memory get valuable. An agent that holds your operating context for years is an asset only if that context provably cannot leak into anyone else’s tenant — or anyone else’s model.
When you want no memory at all
Some work should leave no trace. Run a turn with incognito: true and the platform makes no durable writes for it — no conversation history, no memory entries, nothing added to the knowledge graph. It is a local no-retention mode you invoke per request, not a setting you have to remember to flip back.
Incognito governs monopea’s own persistence — what the platform writes and keeps. The turn is simply never recorded, which is the part of retention monopea controls directly.
Why it matters
No re-briefing
The agent already holds your context. You describe the task, not the backstory.
Compounding, with custody
Memory grows more valuable over time — and stays tenant-isolated, Swiss-stored, and yours to delete.
Opt out per turn
incognito: true runs a turn with no durable writes when the work should not be remembered.
FAQ
Long-term memory, in short
- Is my memory used to train models?
- No. Memory is private to your tenant and is never used to train shared models. It exists solely to make your agent more useful to you.
- Can I see and edit what it remembers?
- Yes. Memory is inspectable, and you can remove anything the agent has retained at any time. For work that should never be retained, run the turn with incognito: true and no durable writes are made.
- How is memory kept isolated from other tenants?
- Two layers: application-level scoping and Postgres row-level security enforced with runtime-minted tenant tokens. Isolation is a database guarantee, not an application convention.
- Where is memory stored?
- At rest in Zurich, Switzerland, with the rest of your persistent data — processed on EU infrastructure like everything else on the platform.
Keep exploring
Swiss residency & routing
Persistent data stored in Zurich, Switzerland; processing on EU infrastructure; and a Swiss model track that keeps inference in Switzerland when you choose it.
Goals & autonomous runs
Hand the agent an objective and it plans and runs the work — on schedules, around the clock — while you steer, stop, and approve.
Skills & routines
Learned, reusable procedures that make repeat work dependable — and never outrun the approval gate.