MrDoing reads code, plans changes, runs tools, fixes tests, and keeps its own memory. No framework, no lock-in, no Python install — single binary, one command to install, works from any project directory.
$ curl -fsSL https://www.doingteam.com/install.sh | bashDrops a wrapper at ~/.local/bin/mrdoing.
$ curl -fsSL https://www.doingteam.com/install.sh | bashSame as macOS — wrapper at ~/.local/bin/mrdoing.
$ irm https://www.doingteam.com/install.ps1 | iexRegisters mrdoing in $PROFILE. Works in any PS window.
Read files, write patches, run bash, grep, run tests — chained automatically toward the goal you asked for.
Remembers conventions across sessions. Project preferences persist in a local MEMORY.md the agent writes to itself.
Live unified-diff rendering with per-tool color, collapsible cards, image paste, and mouse interaction.
Explore / Plan / Review subagents run in parallel for big investigations without bloating your context window.
When a tool errors, you get [Retry] / [Edit & retry] / [Skip] inline — intervene at 1s latency, not after 20s of flailing.
No analytics. No calls home. Your prompts and data never leave your machine unless you point it at a model API.
No. MrDoing ships as a single ~25 MB binary. No Python, no venv, no pip install.
Any Anthropic model (Opus, Sonnet, Haiku) via your own API key, or any Ollama model locally. You pick on first launch and it's remembered.
Run the same install one-liner again. Replaces the binary in place.
Delete ~/.mrdoing (Mac/Linux) or %LOCALAPPDATA%\\MrDoing (Windows) and remove the block between # >>> mrdoing / # <<< mrdoing in your shell profile.
The CLI is closed source for now. Individual contributions land in the DoingTeam community — share tools, share plugins, share prompts.