Kove Model
Rohan 3
The text and reasoning brain of Kove.
Rohan 3 is the model behind every chat, plan, and reply in Kove — fast, well-rounded, and tuned for the way people actually work. It handles daily writing, research, planning, brainstorming, and drives Kove's native tool use: web search, charts, images, media, and documents. It reads what you attach too — images, audio clips, and PDFs all land in the conversation. No need for a button; just ask Rohan.
Kove Model
Kove Image
Generation and editing, side by side.
Kove Image is a pair of sibling models: one composes new images from a prompt in seconds, the other edits an existing image with a natural-language instruction while preserving everything you didn't ask to change. It works on anything living in your chat — generated, searched, or uploaded — and drops results straight into your gallery, ready to download or attach to the next message. It comes in two tiers you choose per image: Standard renders in seconds for a fraction of a cent, while Max steps up to a far more detailed engine with accurate typography and photoreal fidelity — ideal for vision boards, posters, and product shots. Editing has the same two tiers. Billing is per megapixel, so you pay for exactly what you render.
Kove Model
Kove TTS
The voice of Kove.
Kove TTS is the speaking half of the workspace: warm, human prosody with a curated cast of voices — Tessa, Olivia, Clive, and Tyler — that can carry a conversation without the flat 'AI narrator' feel. It powers voice mode, the read-aloud button on every reply, and Kove Podcasts, streaming sentence by sentence so playback starts almost the moment Kove starts talking.
Kove Model
Vine
Print-ready vector design, on demand.
Vine writes SVG from scratch — posters, cards, flyers, invitations, business cards, social posts — powered by Rohan 3. Because the output is true vector, everything is infinitely scalable, tiny in file size, and easy to edit downstream. Vine is where Kove goes when typography matters and pixels don't. It composes with a clear viewBox, native shape tags, and logical groups so the result is clean enough to hand to a designer.