How does the pitch generator work?

It reads your filled-in canvas and produces a 90-second spoken pitch in about two seconds. Five tones × six audiences = thirty distinct combinations. Each weights the nine sections differently and uses a different verbal rhythm. The model runs locally — WebGPU or your own Ollama daemon.

Do I need to write the pitch first?

No. The canvas IS the pitch source. Write the boxes; the generator turns them into a spoken paragraph. You can edit the generated pitch, but you don't have to write one to start.

Can I hear the pitch read aloud?

Yes. The default uses your browser's speech synthesis (free, every browser has it). For a more natural sound there's an optional 80MB Kokoro-82M neural voice download — a one-time, fully offline upgrade.

Why ninety seconds specifically?

The elevator-pitch convention. Most investor first meetings give the founder 30-90 seconds to land the thesis. Generating at the upper bound forces the canvas to fit into a real-world delivery window — if it doesn't, the canvas is too vague.

打字 · 听一听 · 发出去

九个格子，九十秒。带语音路演的精益画布。

把画布敲出来，按下播放，听这段路演。整套都在本地跑。

三个本地 AI 助手追问下一个问题。一段 90 秒的语音 reel，把九个板块变成你能直接递给陌生人的东西。全在你的浏览器里完成。

献给那些宁可听自己的路演、也不想再读一遍计划书的创业者。

开始一份空白画布加载 Airbnb 示例

为什么光打字还不够

读是悄悄的，路演是吵吵的。大多数创始人只读自己的商业模式——从来不去听它。

把自己的生意大声说出来，本身就是一个调试器。屏幕上看着平衡的句子，张口变成十七秒的独白。读起来利落的词，到舌头上就磕巴了。纸上看似稳的「问题」到「解决方案」之间那座桥，到你真的要交付的时候就不见了。在真正重要的那一次之前，你站在某个人面前把那番话说出口之前，你都不会注意到这些。

或者你在周二晚上 11:14 按下一个按钮，听一段从你三十秒前刚写下的九个格子里生成出来的 90 秒语音路演。你听见那个不合适的词。你听见过度解释的那一段。你听见 Founder 风格的版本听起来很有愿景，但 seed-investor 风格的版本却平掉了。你回去，改一改，再按一次播放。

五种语气和六种听众交叉，三十种不同的组合，全在你的笔记本上生成。第一段两秒钟跑完，下一段几乎瞬间。等你真的要站到一群真听众面前，你已经针对三十种听众排练过了。

打字。听一听。发出去。就按这个顺序，每次九十秒。

精益画布在九个格子里做的事。

精益画布是一份单页的商业模式——问题、解决方案、独特价值主张、渠道、客户细分、不公平优势、关键指标、成本结构、收入来源。它用九个创业者真正必须回答的问题，替代一份 30 页的商业计划书，用来判断这个想法是否成立。

Totally Lean 在你的浏览器里跑这份画布。没有账户，没有服务器，数据留在你本地的存储里。三个本地 AI 助手帮你填写；一段 90 秒的语音路演生成器，把画布变成你能直接递给陌生人的东西。

点击"开始"，几秒钟内就能编辑——不注册、不留邮箱、没有团队工作区。想让别人加入，就把 URL 分享出去。想留一份副本，就导出 JSON。

其他创业者把同一份画布套用的方式。

同样的九个格子，针对不同时刻被裁切。选一份贴合你当下处境的。

Frequently asked

How does the pitch generator work?: It reads your filled-in canvas and produces a 90-second spoken pitch in about two seconds. Five tones × six audiences = thirty distinct combinations. Each weights the nine sections differently and uses a different verbal rhythm. The model runs locally — WebGPU or your own Ollama daemon.
Do I need to write the pitch first?: No. The canvas IS the pitch source. Write the boxes; the generator turns them into a spoken paragraph. You can edit the generated pitch, but you don't have to write one to start.
Can I hear the pitch read aloud?: Yes. The default uses your browser's speech synthesis (free, every browser has it). For a more natural sound there's an optional 80MB Kokoro-82M neural voice download — a one-time, fully offline upgrade.
Why ninety seconds specifically?: The elevator-pitch convention. Most investor first meetings give the founder 30-90 seconds to land the thesis. Generating at the upper bound forces the canvas to fit into a real-world delivery window — if it doesn't, the canvas is too vague.

九个格子，九十秒。带语音路演的精益画布。

不注册，没有服务器，没有邮箱。打开编辑器，开始动笔。

开始一份空白画布加载 Airbnb 示例