Train your first model
in five minutes.

Quickstart

The fastest way to train a model on Axondyne is the Axon CLI. Install it with a single command:

# macOS / Linux
curl -fsSL https://get.axondyne.com/install.sh | sh

# verify
axon --version

Authenticate with your API key (find it in the console under Settings → API keys):

axon auth login --key ax_live_xxxxxxxxxxxxxxxx

1. Prepare your data

Axon expects a JSONL file where each line is a training example. For an instruction fine-tune:

{"messages": [
  {"role": "user", "content": "Summarize this ticket..."},
  {"role": "assistant", "content": "The customer reports..."}
]}

2. Launch a fine-tune

axon train \
  --base ax-mini-7b \
  --data ./support.jsonl \
  --epochs 3 \
  --eval-gate 0.90

The --eval-gate flag blocks the model from shipping unless it passes your evaluation threshold. Watch progress live:

axon runs watch latest

3. Deploy

axon deploy ax-support-7b --target vpc --region us-w2

That's it — your model is live behind a private endpoint. Call it from the API below.

---

CLI reference

API reference

Every CLI action has an HTTPS equivalent. Authenticate with a bearer token:

curl https://api.axondyne.com/v1/chat/completions \
  -H "Authorization: Bearer $AXONDYNE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ax-support-7b",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Responses follow a standard OpenAI-compatible schema, so most existing SDKs work by just swapping the base URL. See the full reference for streaming, tool use, and batch endpoints.

Tip: set AXON_BASE_URL=https://api.axondyne.com/v1 and most client libraries will work unchanged.