Skip to content

LOCAL AI CANVAS · AGENT-DRIVABLE

A Lean Canvas with a real local AI coach. Your idea stays on your laptop.

Nine boxes. One page. Three offline agents that interrogate the canvas. And every AI tool you already use can read it.

Onboarding, Critic, and Explorer agents run in your browser or on your Ollama daemon. The DOM contract and .leancanvas file format are published — agents like Claude, Codex, or Cursor can read and operate your canvas without anyone else seeing it.

For founders pairing with AI — without handing over the idea.

Ce qu'un Lean Canvas fait en neuf cases.

Un Lean Canvas est un modèle d'affaires sur une page — Problème, Solution, Proposition de Valeur Unique, Canaux, Segments de Clients, Avantage Concurrentiel Injuste, Indicateurs Clés, Structure de Coûts, Flux de Revenus. Il remplace un business plan de 30 pages par les neuf questions auxquelles un founder doit vraiment répondre pour savoir si l'idée tient.

Totally Lean fait tourner le canvas dans ton navigateur. Pas de compte. Pas de serveur. Les données vivent dans ton stockage local. Trois agents IA locaux t'aident à le remplir ; un générateur de pitch parlé de 90 secondes transforme le canvas en quelque chose que tu peux tendre à un inconnu.

Quand tu cliques sur Démarrer, tu édites en quelques secondes — pas d'inscription, pas d'email, pas d'espace d'équipe. Partage via URL quand tu veux faire entrer quelqu'un. Exporte en JSON quand tu veux une copie.

D'autres façons dont les founders cadrent le même canvas.

Les mêmes neuf cases, cadrées pour des moments différents. Choisis celle qui correspond à là où tu en es maintenant.

Frequently asked

Can Claude, Codex, or Cursor operate this canvas for me?
Yes. Totally Lean publishes an operating manual for AI agents at /agent.txt — a public document describing the canvas data model, stable DOM selectors, keyboard shortcuts, and rules agents should follow. Any browser-driving agent (Claude for Chrome, Codex Chrome extension, Operator-class tools) can read the manual and operate the canvas through documented affordances. The agent talks to your browser, not to us — your canvas never reaches a server. The .leancanvas JSON file format gives agents a structured read/write channel as well.
What do the three local AI agents do? Why three?
Three different conversations help at different stages of writing a canvas. Onboarding interviews you when the canvas is blank. Critic challenges what you've written — weak Problem statements, vague Customer Segments, hand-wavy Unfair Advantages. Explorer brainstorms adjacencies you haven't considered. Each holds a structured written conversation about the canvas; none overwrites it directly. All three run on your machine — WebGPU in your browser or your own Ollama daemon.
Is the AI really local? Where does the model actually run?
Two paths. (1) WebGPU — model weights download once and cache in your browser; inference runs on your GPU. (2) Ollama — your own daemon running locally; the browser talks to it via localhost. Neither sends canvas content to any server we operate. The default behavior is fully offline.
Can I read or write the .leancanvas file with my own tools?
Yes. The format is documented at /file-format with a JSON Schema (Draft 2020-12). Open it in a text editor, parse it with any JSON library, edit it programmatically — versioned, stable, agent-friendly. The format is the same shape the editor uses internally; no proprietary lock-in layer between what you see and what gets exported. If Totally Lean disappears, your canvases stay readable forever.

A Lean Canvas with a real local AI coach. Your idea stays on your laptop.

Pas d'inscription, pas de serveur, pas d'email. Ouvre l'éditeur et commence à griffonner.