Quickstart

Install the fn CLI, scaffold an Agentic Function, run it, evaluate it, and deploy it — in about five minutes.

1. Install

curl -fsSL fnai.dev/install.sh | sh

Confirm the binary is on your PATH:

fn --version
# fn 1.0.0

2. Scaffold an Agentic Function

fn init classify --template classify

This writes an fn.yml describing a typed Agentic Function for classification:

name: classify
version: 1
model: claude-opus-4-8

input:
  ticket: file

output:
  category: string
  priority: enum[low, high]
  confidence: number

3. Run it locally

fn run classify --input ticket.pdf

{
  "category": "billing",
  "priority": "high",
  "confidence": 0.94
}

The metadata line below the output reports the version, latency, cost, and the last eval score so you always know what you just ran.

4. Evaluate

Create a dataset of real inputs and expected outputs, one JSON object per line:

{"input": "ticket-001.pdf", "expected": {"category": "billing"}}
{"input": "ticket-002.pdf", "expected": {"category": "support"}}

Then grade against it:

fn eval classify --dataset tickets.jsonl --threshold 0.95

A score below the threshold exits non-zero, so you can drop this straight into CI. See the Evaluation guide for datasets and gates.

5. Deploy

fn deploy classify --env production

You now have a versioned endpoint. Call it like any other function, from your code:

Or hit the endpoint directly over HTTP:

curl -X POST https://api.fnai.dev/v1/run/classify \
  -H "Authorization: Bearer $FN_KEY" \
  -H "Content-Type: application/json" \
  -d '{"input": {"ticket": "…"}}'

Next steps

• Learn the full CLI
• Read the fn.yml spec
• Set up deployment environments