Bruto
1 comment
Bruto is a local-only task board for coding with AI. Tasks live as notes in a .bruto/workspace.json file inside your project folder, so you and every AI agent you use work from the same source. Write a task with files, links, and screenshots attached, copy the context with one key (Q for the selection, W for the selection with everything it points to, E for the whole project), and paste it into Claude, ChatGPT, Gemini, or DeepSeek. If the agent can edit files, it answers inside the note and marks it for review. What the AI answers stays in the note with all its context, so if something is wrong you write what and send it back. Standing rules marked as Loop go into every copy. The structure view draws your project as it is — pages with components for React, Vue, Svelte, Next.js, Nuxt, SvelteKit, Astro, and Angular; API resources with endpoints for .NET, NestJS, Express, Fastify, Next.js, FastAPI, Flask, and Spring; tables with keys and policies for PostgreSQL, Supabase, Prisma, and Drizzle. An MCP server lets Claude Code and other agents list, read, and answer notes without editing JSON. Notes can block or wait on tasks in other projects. Keyboard first, nine languages, two themes, works offline, installs as a PWM. No server, no accounts, no cookies. Free and open source. Built by Juan Llinares.
Meet the team
What people who use it say
Nobody has reviewed Bruto yet. If you have used it, you are the one who can say whether it holds up.
Write the first review
Sign in to comment
The gap this fills is real: when you work with an AI on a project, the tasks live in one place, the context in another, and the AI's answers disappear into a chat history you will never search. Bruto keeps tasks as notes in a file inside your project, so the board, the context, and the AI's responses are all in one place that survives switching agents.
The one-key context copy is what convinced me. Press Q and it copies exactly what the AI needs to understand that task — the note, the attached files, the links — as clean Markdown with token counts shown before you paste. No more explaining the same architecture three times to three different agents. I have not used the MCP server yet. The first thing I would check is whether an agent with file access can actually answer inside the note and mark it for review without breaking the JSON, or whether the merge handling falls apart when you and the agent both edit the file between reads.