Open Sourcing the Atuin AI Server
Sign in<br>Subscribe
Atuin AI is a fast terminal-focused agent right in your shell. It starts instantly and comes with all the agentic tools needed to help with work in your terminal.
0:00
/0:30
1×
Atuin offers a generous usage budget for free for all users, and we don't store your AI conversations, but we recognize some people would prefer to keep tighter tabs on their data, especially data that originates from terminal usage.<br>To that end, we're happy to be open sourcing the Atuin AI server.<br>The Atuin AI Server<br>The Atuin AI server can be found on GitHub at atuinsh/atuin-ai-server; it's based on atuinsh/atuin-ai-core, the same library we use for the production Atuin AI server.<br>The Atuin AI server currently supports any OpenAI-compatible, chat completions-style endpoint. For local models, this includes Ollama, vLLM, LM Studio, llama.cpp, and LiteLLM, among others. You can also use OpenAI-compatible web services, like OpenRouter.<br>Getting Started<br>After cloning the repository, copy the example config file, config.example.toml, to config.toml. Follow the configuration section of the readme to set up your instance.<br>Here's a very basic example of an Ollama-based setup:<br>port = 8080<br>endpoint = "http://localhost:11434/v1" # or host.docker.internal<br>api_key = "ollama"
default_model = "llama31"
[request.body]<br>stream_options = { include_usage = true }
[[models]]<br>alias = "llama31"<br>name = "Llama 3.1 70b"<br>description = "Ollama Llama 3.1 70b"<br>model = "llama3.1:70b"
[[models]]<br>alias = "gemma4"<br>name = "Gemma 4 r4b"<br>description = "Ollama Gemma 4 - Effective 4b"<br>model = "gemma4:e4b" See the repository readme for more setup details, including configuring server-side tools, like web search and web content scraping.<br>Once done, you can start the server one of two ways:<br>Running from Source<br>If you have Erlang, Elixir, and Gleam installed, you can run the server natively:<br>mix deps.get<br>mix run --no-haltIf your config.toml specifies API keys via environment variables, remember to set them when you start the server.<br>Running with Docker<br>To run the server with docker, run the following:<br>docker run \<br>-v ./config.toml:/etc/atuin-ai/config.toml \<br>-p 8080:8080 \<br>ghcr.io/atuinsh/atuin-ai-server:latestIf you're running via Docker and want the Atuin AI server to connect with a local LLM service running on the host, like Ollama, use host.docker.internal as the endpoint instead of localhost (which would resolve to the container's own loopback interface).<br>Configuring Atuin AI<br>Once your server is running, you can configure Atuin AI to connect to it by setting the endpoint config:<br>[ai]<br>endpoint = "http://localhost:8080"
Read more
Atuin v18.13 – better search, a PTY proxy, and AI for your shell
A new release is out! v18.13 is probably the biggest set of changes we have released in a good while, read on to find out more.
Much faster and better search with the daemon
The daemon has existed for a long time, and has been marked as "experimental&
Custom Keybindings for the Atuin TUI
One of the longest-standing feature requests for Atuin has finally landed: full custom keybinding support for the search TUI. No more being stuck with our defaults, no more wishing ctrl-d did something else. Your TUI, your rules.
Atuin Devlog: Desktop AI, UI + performance improvements, SSH fixes, and more
Welcome to the first installment of a monthly post detailing what's new in Atuin Desktop. This post is a little unique as we're covering a bunch of stuff from both 2025, and the first few weeks of 2026.
AI Assistant
Agents changed how many engineers write
Introducing the New Runbook Execution Engine
We're excited to announce a major architectural improvement to Atuin Desktop: a completely redesigned runbook execution engine.
This is a huge change, the first big step toward making runbooks a core automation primitive.
If you’ve ever hit flaky context, disappearing state, or inconsistent execution, this release fixes
Powered by Ghost