Chat-first, orb at the centre
The whole product is one conversation with a living orb that breathes with the work — idle, listening, thinking, speaking — so you always know what your agent is doing.
Kavvo is a skill-first AI agent that lives in chat. Talk or type — it listens, thinks, browses, remembers, and quietly creates its own skills in the background so it gets better at your work every day. You bring the API keys; Kavvo brings the agent.
Free forever · Bring your own API key · Runs on your machine or your server
No plugin stores to browse, no prompt gymnastics. Kavvo is built around a single conversation — everything else happens around it.
The whole product is one conversation with a living orb that breathes with the work — idle, listening, thinking, speaking — so you always know what your agent is doing.
A real listen → think → speak flow. Start talking and Kavvo listens; it thinks in the open, then speaks back. Interrupt it mid-sentence and it stops and listens again.
Kavvo drives a real-human browser — opening pages, reading them, filling forms, and attaching to your own browser session when a task needs your logins.
Long-term memory keeps your projects, preferences and facts straight across sessions — guided by a constitution you write and can edit any time.
Morning briefings, weekly reviews, overnight delegations. Kavvo prepares the summary before you ask, and reports what it did while you were away.
Every action — a page opened, a skill written, a memory saved — streams into a live feed beside the chat. Nothing happens off-screen or without a trace.
Paste an API key from a model provider you already use. The key stays on your machine or server and talks to the provider directly — Kavvo never sits in the middle.
Kavvo fetches the provider's real, current model list automatically. Choose the exact model you want, or leave it on Auto and let Kavvo route each job sensibly.
Type or speak like you would to a sharp colleague. Kavvo plans, browses, remembers, and builds the skills it needs — then shows you everything it did.
Most AI tools hand you a store of add-ons to install and babysit. Kavvo works the other way: when a task needs a capability it doesn't have, it creates that skill itself, keeps it in the background, and reuses it next time. Your only job is the conversation.
Kavvo is bring-your-own-key by design: your provider keys stay on your machine or server, and your conversations go to the provider you chose — not to us. And when something should not be remembered, the privacy guardian can purge a whole topic from memory and hand you a receipt of exactly what was removed.
Kavvo itself costs nothing. You pay only your model provider, directly, for the tokens you actually use — no markups, no seats, no tiers of features held hostage.
The full v1 app, run locally or on your own server.
A managed home for Kavvo, for people who don't want to run a server.
Yes. The Kavvo app is free and open source under the MIT License — download it, run it, read the code, change it. The only cost in the picture is your own model usage: because Kavvo is bring-your-own-key, your provider bills you directly for tokens, at their prices, with nothing added by us.
Any model from a provider whose API key you connect. When you add a key, Kavvo fetches that provider's real, current model list automatically — you pick the exact model you want for a chat, or choose Auto and let Kavvo decide per task. Switch models any time; your memory and skills carry over.
On the machine or server where you run Kavvo. Keys are used locally to call your provider directly — they are never sent to us, and there is no Kavvo account required to run the app. Memory also lives with your install, under your control, and can be exported or purged.
A skill is a capability Kavvo writes for itself when a task needs one — a repeatable way of doing something, saved and reused. Unlike plugin stores, skills are background machinery: Kavvo creates, tests and improves them on its own. You will see them mentioned in the activity stream, but you never install or manage them.
Yes — that is the privacy guardian's job. Tell Kavvo to purge a topic and it removes that material from its memory and gives you a receipt listing what was deleted, so "forgotten" is verifiable rather than a promise.
Both. A public preview instance runs at the "Open Kavvo" link — open it, connect a key, and talk. For keeps, run Kavvo on your own computer with the Windows bundle, or on any small server you control. A managed hosted tier is planned for later.
Open the live preview, connect your key, and have your first real conversation with Kavvo.