Tacuri AICode with an agent that runs, tests and fixes.
It reads your project, works in the terminal, edits files and opens the site it started to check it. It reviews your changes, watches your PRs' CI, and you approve every change.
With the models you already know, included in your account
- DeepSeek
- Claude
- ChatGPT
- Kimi
- ZGLM
- Qwen
Everything it takes to finish, not just to start
From request to green tests
It reads the code, finds where everything is defined with the language server, edits, runs the tests and, if they fail, reads the error and fixes it. It follows your project's AGENTS.md: how to build, what not to touch and your style.
Several agents, each on its own branch
Sub-agents with their own role (reviewer, tests, research) or several tasks in parallel, each in its own copy of the repository. Message one halfway through, or stop it and keep what it had done.
Reviews, opens the PR and watches CI
/revisar and /revisar-seguridad put the bugs on a card, from most to least serious. When it opens a PR it keeps watching: if a check fails, it conflicts with the base branch or someone comments, it fixes it and pushes again.
The board on your desk
Arduino, ESP32, Raspberry Pi Pico, STM32, micro:bit and MicroPython. It sees what's plugged in, compiles, flashes and reads the serial monitor to check it booted. Flashing always asks for permission.
Give it a job
Example tasks: what you ask for and what it delivers.
“The payments test has been failing since yesterday. Fix it.”
Done. The total was rounding before adding taxes and lost a cent. I changed it in total.ts and all 48 tests pass.
Everything it does
The terminal and your files
It runs commands, installs dependencies and edits code. Whatever it needs (Git, Python, ripgrep…) it installs after asking.
Every change, in plain sight
The change list and git built in: diffs, branches and commits. “Restore” takes the conversation and the files back to any point.
A browser for testing
It opens the site it started, checks it on mobile and desktop, clicks, reads the console and the network log, and views it in dark mode.
Several agents at once
Sub-agents with their own role or tasks in parallel, each in its own copy of the repository, and it lets you know when they're done.
Boards and firmware
It detects the plugged-in board, compiles, flashes and reads the serial monitor. Each board's tools download on their own.
The tools you already use
Each project's skills, agents and / commands, two marketplaces, and connectors like GitHub, Linear, Sentry or Notion.
Clear permissions
Manual asks before every sensitive action; Auto carries on by itself. Commands run in a sandbox and everything is logged.
Notebooks and your code
It reads Jupyter notebooks cell by cell and edits them; it checks TOML, JSON, Python and TypeScript before calling it done.
On your server or in the cloud
The same agent on your server or in your cloud: it keeps working with your PC off and you check on it from your phone.
The best models, or yours
The ones in your account with nothing to set up, your keys encrypted on your server, or models on your computer with Ollama.
- Your keys· encrypted on your server
- DeepSeek· included
- Claude· included
- ChatGPT· included
- Kimi· included
- GLM· included
- Qwen· included
- Ollama· on your computer, offline
- Gemini
- Groq
- Your provider· OpenAI-compatible
Ask Tacuri AI.
With your models, on your PC, on the web, on your phone and in Chrome.