docent v1.0.59 · open source · MIT
A coding agent for your terminal. On open models.
docent is an open-source AI coding agent for your terminal that runs on open models. It writes code as diffs, checks it against your language servers and build, fixes its own errors, and runs the commands once you say so.
$curl -fsSL https://smartloop.ai/install | shfree · macOS, Linux, Windows · no account
install → pick an open model → a Rust program that builds and runs · 1:33
What it does
Ask for a change. Get a build that passes.
- Writes code. New files whole, changes as small edits, each shown as a diff as it streams.
- Checks its work. Runs your language servers and build, and hands errors back to the model until it passes.
- Runs the work. Commits, installs and tests, once you say so.
- Open models. No vendor lock-in: pick the model, and run it locally so your code stays private.
How it works
What happens when you ask for a change.
- 01
It reads the project.
The file tree, the files your prompt is about and the git state (branch, remotes, what changed), plus your OS, shell, package managers and toolchains, so it answers from your code and uses brew on a Mac and apt-get on Ubuntu.
- 02
You see diffs as they stream.
Each file shows as numbered hunks, additions in green and removals in red, and is written into the folder as soon as it is done. Changes to files that already exist ask first.
- 03
It checks the build and fixes it.
A fast check for the kind of project runs after every change. Errors go back to the model, round after round, until the check passes.
- 04
Commands wait for your OK.
Commits, installs and test runs come back as shell commands. Enter runs them, Tab approves the rest of the session, Esc skips. Their output goes back to the agent, so a failing test gets fixed next.
- 05
Long runs go to the background.
Dev servers and watchers keep running under the prompt while you keep chatting. /tasks shows them and /stop ends them.
More in the coding guide.
Works with your stack
Checked by the tools you already use.
docent reads code in any language. For these it also knows the language server and the build check, so errors are caught and fixed before you see them.
rust-analyzercargo checkgoplsgo buildtypescript-language-servertsc --noEmitpyrightpy_compileclangdyour build—bash -nOpen models
Your model, your machine.
Local, on your GPU
Small open models such as Qwen, Gemma and Ministral run on your computer, through Metal on macOS and Vulkan on Linux and Windows. Your code stays on the machine.
Bigger open models, when you need them
Sign in to add larger open models such as MiniMax M3, GLM and the bigger Qwen models. Switch any time with /models; no vendor lock-in.
macOS, Linux, Windows
Signed builds for Apple Silicon Macs, Linux on x86_64 and arm64, and Windows on x86_64. One command installs it; no account needed.
More than code
An agent harness, not just an autocomplete.
- Chat with your files. Drop a PDF, Office file, CSV, text file or image on the terminal and ask about it. Documents are indexed on your machine.
- Web search. Sign in with /login for AI search, so answers can cite current docs and library APIs, with their sources.
- MCP servers. Connect tools with /mcp add <url> and docent can call them while it works.
- Scriptable. Pipe a plain answer into your scripts with --plain, or run docent over SSH.
Every command and shortcut is in the command-line reference; see also documents and MCP servers.
Get started
From install to a passing build.
Three things to type. docent does the rest.
- 01
Install
One line puts docent on your PATH. macOS, Linux and Windows.
curl -fsSL https://smartloop.ai/install | sh - 02
Open a project
The first run fetches what it needs. Type /models to pick an open model, like MiniMax M3.
cd my-app && docent - 03
Ask for a change
docent writes the diff, builds it, fixes what breaks, and asks before it runs anything.
> add a CLI that asks for a name and greets it
FAQ
Questions, answered.
What is docent?
docent is an open-source AI coding agent for the terminal, made by Smartloop. Open it in a project, ask for a change, and it writes the code as diffs, checks it with your language servers and build, fixes its own errors, and runs commands like commits and tests once you approve them. It also answers questions about your documents and the web.
Is docent free and open source?
Yes. docent is MIT licensed, with the source on GitHub, and it is free to use with no account. Signing in is optional: it unlocks AI web search and the cloud models, which draw on a monthly allowance.
Which models does docent use?
Open models. Small ones such as Qwen, Gemma and Ministral run on your own machine, and larger open models such as MiniMax M3, GLM and the bigger Qwen models are available in the cloud when you sign in. Pick and switch models with /models; the one that suits your machine is marked.
Does my code leave my machine?
Not with local models: docent runs the model and its agent on your computer, with GPU acceleration through Metal on macOS and Vulkan on Linux and Windows. Cloud models are marked in /models, and web search sends only the search query, never your files.
Which languages does docent support?
Any language it can read, with built-in checks for Rust (cargo check), Go (go build), TypeScript and JavaScript (tsc --noEmit), Python (py_compile) and shell scripts (bash -n). It also uses your language servers, such as rust-analyzer, gopls, typescript-language-server, pyright and clangd, to catch errors.
Which platforms does it run on?
macOS on Apple Silicon (signed and notarized), Linux on x86_64 and arm64, and Windows on x86_64 (signed). Install with one command: curl -fsSL https://smartloop.ai/install | sh on macOS and Linux, or irm https://smartloop.ai/install.ps1 | iex in PowerShell.
Will it run commands without asking?
No. Files it creates are written straight away, but changes to existing files ask first, and every shell command (a commit, an install, a test run) waits for you to press Enter. Press Tab to approve the rest of a session, or Esc to skip.