Skip to content

Install

Release binary

Download the archive for your OS from GitHub Releases (tags v*). Available targets: Linux x86_64, Windows x86_64, macOS x86_64 + aarch64. Put attx / attx.exe on your PATH.

From source

git clone https://github.com/kaecho/attx.git
cd attx
cargo build --release
./target/release/attx --help
cargo install --path .   # optional

Requires a recent stable Rust (edition 2024). No nightly features, no MSRV pin.

LLM config — two paths

Install the Skill, then ask the agent to set up attx. It walks endpoint → key → model → languages and writes setting.toml without echoing the key. See Agents.

B. Manual

cp setting.example.toml setting.toml

[llm]
default_client = "main"

[[llm.clients]]
name = "main"
provider_type = "openai"          # OpenAI-compatible Chat Completions
base_url = "https://api.example.com/v1"
api_key = "YOUR_API_KEY"
model = "your-model"
timeout = 600                     # seconds, per request

[translation]
worker_count = 8       # parallel HTTP batches
rpm = 60               # global rate limit per minute (0 = unlimited)
retry_count = 3
retry_delay = 2
batch_chars = 2500     # max source chars per batch
max_context_items = 6  # max units per batch

Then verify:

attx doctor --ping

doctor checks config, lists the built-in adapters and saved profiles; --ping also sends one tiny request to the LLM. Machine-readable form: attx doctor --json.

Config search order

--config <path> → $ATTX_HOME/setting.toml → ./setting.toml.

  • ATTX_HOME also holds your saved profiles and learned experience: $ATTX_HOME/profiles/, $ATTX_HOME/knowledge/.
  • Without ATTX_HOME, the platform config dir is used (~/.config/attx/ on Linux).
  • setting.toml is gitignored — never commit API keys.
  • --client <name> switches to a non-default [[llm.clients]] entry for one invocation.

Verify the install

attx formats                 # list of built-in adapters as JSON
attx detect --input <file>   # which adapter claims your input
attx --help