Independence is structural.

The canvas, not the chat box

Wavn is a block-based canvas. Work lives as blocks, text, documents, images, that you can structure into departments, folders, and canvases. Conversations (waves) attach to specific blocks rather than floating in a single linear thread. This is the foundation: structure first, AI where it earns its place.

Three minds, each in its lane

Every wave can call three independent minds, Claude, Gemini, and Cohere. Each reply is labelled by mind. You can ask one, ask all, or have them debate. No mind is privileged as the source of truth, and OpenAI is never used.

This isn't orchestration for its own sake. Diversity of model means diversity of failure: where one mind is confidently wrong, the others rarely share the exact blind spot, and Converge makes that divergence visible instead of averaging it away.

Why no single vendor

Single-vendor dependency is a risk we designed out. It concentrates failure, it narrows the range of views your decisions are exposed to, and it ties your workspace to one provider's availability, pricing, and policy. Three independent minds remove that concentration. For teams with their own constraints on which providers may process their material, the architecture is also a practical answer.

Decisions as durable objects

When you Decide, the outcome becomes a first-class object on the canvas, a resolved decision block carrying its reasoning, dissent, and open questions. Memory indexes these so they can be recalled into future waves, revisit dates re-argue them against what changed, and the Journey report can replay a project's whole history in order. The result is a workspace where conclusions accumulate rather than scroll away.

Decisions as inputs to real work

A locked decision isn't the end of the pipeline, it's the input to the rest of it. Deliverables (documents, decks, styled spreadsheets, web pages) generate directly from the decision and its reasoning. Integrations push actions into GitHub, Linear, Asana, Jira, and Slack. And through MCP, any AI client, most naturally Claude, can read a decision, build against it, ask its questions back on the canvas, and deliver the result home. The decision record is the API.

The Guide as a state machine

The road on every canvas (Idea, Argued, Decided, Set up, Spec, Built, Scheduled) isn't decoration, it's derived state. Wavn reads what has actually happened on the canvas, which arguments ran, whether a decision locked, what got built, and lights the stage you're truly at, opening exactly one next door at a time. Guidance is computed from the work, not from a tutorial script.

Honest about maturity

Wavn is early. Some of what a mature platform needs, formal certifications, residency options, deeper enterprise controls, is on the roadmap rather than shipped, and we say so plainly. See Security and Trust.