dothething logo

dothething

an autonomous AI agent: you describe the thing, it does the thing.

ShellAI AgentsShellBSD-3-Clause 943 178Health 80
Website GitHub

What is it?

What it is

A local AI agent that accepts plain‑English tasks, plans work, and automatically uses web search, browser automation, file I/O, code execution, email, and custom skills to produce a final report or perform actions.

Why it exists

To let users delegate complex, multi‑step research and automation tasks to an AI that can run unattended, persist progress, and manage costs, without relying on a remote server for every interaction.

Who should use it

Technical users who want to automate complex, multi-step tasks involving web interaction, file operations, and code execution, and are comfortable with the command line and managing API keys.

Who should avoid it

Users on Windows, those preferring GUIs, and anyone unwilling to set up required API keys (OpenRouter) or optional services (like 2Captcha for captcha solving).

How it works

A quick walkthrough in plain English

How dothething works

Step 1 of 3

You interact with it

Open dothething, send a request, or connect it to your stack.

Features

Local AI agent that can plan, browse, edit files, run shell commands, and execute code
Web search using a local SearXNG instance with multiple engines
Browser automation via Notte and Camoufox, including captcha solving
Email integration through AgentMail
Clipboard access for text and images
Mid‑task interactive input and queued commands
Cost and token tracking via OpenRouter with prompt caching
Skill system for custom procedures (inline or delegated)
MCP server integration for additional tools
Orchestrator mode for running multiple agents concurrently
Quick mode for rapid one‑shot tasks
Pipe mode for stdout‑only output
Desktop and email notifications
Cost limits and resume capability
Support for multiple models and role overrides

Advantages

  • Open source (BSD-3-Clause)
  • Active Shell ecosystem
  • Self-hosted deployment options

Disadvantages

  • Requires operational ownership for self-hosted setups
  • Community support varies by project maturity

Installation

native

git clone https://github.com/fluffypony/dothething.git
cd dothething
./dtt.sh --prompt "your task"

FAQ

How do I set up dothething initially?

Clone the repository, install dependencies, and provide an OpenRouter API key during first run. The agent will prompt for keys or read them from your shell environment.

What models does dothething use by default?

The main agent uses Claude Fable 5, worker tasks use Gemini 3.5 Flash, browser tasks use Claude Sonnet 4.6, and the oracle uses GPT-5.5 by default.

How does it handle browser automation with captchas?

It uses Notte with Camoufox for browsing and automatically solves captchas via the 2Captcha API key (optional). You can provide the key during setup or via environment variables.

Can I resume a previously interrupted task?

Yes, use the `--resume` flag with the thread ID saved in `~/.dtt/threads/` to continue from where you left off.

What are the system requirements?

Requires macOS or Linux, Python 3.11+, an OpenRouter API key, and optionally 2Captcha/AgentMail API keys. Linux users need `wl-clipboard` or `xclip` for clipboard support.

How does dothething manage costs?

It tracks token usage and costs via OpenRouter. You can set a budget limit with `--max-cost` to stop the task when spending reaches your specified amount.

Loading documentation…
View on GitHub

Featured in Videos

YouTube tutorials and walkthroughs for dothething

Alternatives

Similar projects ranked by category, topics, and text overlap.

Compare
dothething | MushyBook