AI in the Terminal

ait is a terminal user interface for interacting with several generative large language models from multiple providers. It uses the genai crate to communicate with the model providers. The TUI is built using the ratatui crate.

ait2.mov

Installation

Installation requires cargo to be installed.

cargo install ait

Manual installation

Clone this repository and cd to the ait directory and run the application using:

cargo run

Install the application by running:

cargo install --force --path .

The binary name is ait.

Binaries are also available for download under Releases.

Usage

The chat interface is modal and starts in the 'normal' mode. By pressing the i key text can be input into the text area. More information can be found by pressing the ? key. To submit queries to the model providers, you either need to obtain an API key and set the appropriate environment variable OR you need a running Ollama instance on http://localhost:11434.

To start the TUI simply run

ait

If you want to provide a custom system prompt, it can be achieved like this:

ait --system-prompt "You are a helpful, friendly assistant."

If you want to add context to your conversation, use the --context argument.

ait --context my_file.txt

ait can also read from stdin to add context:

cat my_file.txt | ait

A powerful pattern is to use a text serializer such as yek and use this as context input:

yek my_file.txt | ait

Or serialize all file in a directory and add as context:

yek my_dir | ait

Chat history

Chat history is stored as a sqlite database (facilitated by the rusqlite crate) in the platform's standard data directory:

• macOS: ~/Library/Application Support/ait/chats.db
• Linux: ~/.local/share/ait/chats.db
• Windows: %APPDATA%\ait\chats.db

In addition, ait will store a log of the latest chat in the platform's cache directory:

• macOS: ~/Library/Caches/ait/latest-chat.log
• Linux: ~/.cache/ait/latest-chat.log
• Windows: %LOCALAPPDATA%\ait\latest-chat.log

I'm probably the only one using this tool but for users of ait version 0.5.1 and earlier, to keep your old database, simply copy it from the previous location:

cp ~/.cache/ait/chats.db <new platform specific location according to list above>

Configuration

AIT can be configured via a config.toml file. Please refer to this file for a minimal example.

This file should be stored in the platform specific location:

• Linux: $XDG_CONFIG_HOME or $HOME/.config/ait/config.toml, e.g., /home/alice/.config/ait/config.toml
• macOS: $HOME/Library/Application Support/ait, e.g., /Users/Alice/Library/Application Support/ait/config.toml
• Windows: {FOLDERID_RoamingAppData}\ait\config.toml, e.g., C:\Users\Alice\AppData\Roaming\ait\config.toml

MCP (Model Context Protocol) servers

ait supports the Model Context Protocol for extending the assistant with external tools (file access, search, weather, and so on). MCP servers are declared in config.toml under [mcp.servers.<id>] and connect automatically on startup when enabled.

Configuration

Each server has a stable id (the TOML table key) and uses either the stdio transport (a child process) or the http transport (a remote URL). Setting both command and url, or neither, is a configuration error.

[mcp.servers.filesystem]
name = "Filesystem"
enabled = true
command = "npx"
args = ["-y", "@modelcontextprotocol/server-filesystem", "/tmp"]
env = {}

[mcp.servers.weather]
name = "Weather"
enabled = true
url = "https://weather.example.com/mcp"
api_key = "your-secret-key"
headers = { "X-Custom-Header" = "value" }

Field	Description
name	Optional human-readable display name. Defaults to the server id.
enabled	Whether the server connects automatically at startup. Defaults to true.
command	Command to spawn for a stdio server (e.g. npx, uvx).
args	Arguments passed to the command.
env	Extra environment variables for the spawned process (where stdio server secrets such as API keys go).
url	URL of a streamable-http MCP server.
api_key	API key sent as Authorization: Bearer <api_key> on every request to an http server.
headers	Extra HTTP headers for an http server (e.g. X-API-Key).

Secrets

Both transports support environment variable expansion, so you never need to store API keys or other secrets verbatim in the config file. References are expanded at connect time; placeholders stay on disk.

• ${VAR} → value of VAR (error if unset)
• ${VAR:-default} → value of VAR, or default if unset
• $VAR → value of VAR (error if unset)
• $$ — a literal $

[mcp.servers.kagi]
command = "uvx"
args = ["kagimcp"]
env = { KAGI_API_KEY = "${KAGI_API_KEY}" }   # set KAGI_API_KEY in your shell

An unset variable with no default is a hard error, so a missing secret fails loudly instead of silently sending an empty value to a server.

Managing servers in-app

• Press S in normal mode to open the server management view.
• Use j/k or Up/Down to move through the configured servers.
• Press Space to toggle a server's enabled state, connecting or disconnecting it live.
• Press Esc, q, or S to return to normal mode.

The footer shows a summary of MCP server status (ready, connecting, failed) and the connected tools available to the assistant.

How tools are used

When you submit a message, the assistant can call any tool exposed by the currently-connected MCP servers. The model decides when a tool is needed; each call is executed immediately and the result is fed back into the conversation so the model can continue. Tool calls and results are shown inline in the thinking trace, so you can watch what the assistant is doing. The conversation supports up to 12 tool-calling rounds per response to prevent runaway loops.

Example servers

A few well-known MCP servers to try:

[mcp.servers.filesystem]
name = "Filesystem"
command = "npx"
args = ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/sandbox"]

[mcp.servers.everything]
name = "Everything (test)"
command = "npx"
args = ["-y", "@modelcontextprotocol/server-everything"]