CLI Quickstart

Ship your first agent to savine.in in five minutes. This guide assumes zero prior setup.

1. Install

npm install -g savine-cli
savine --version

(Also works with bun install -g savine-cli, pnpm add -g savine-cli, or yarn global add savine-cli. For no-install, one-off use: npx savine-cli ….)

Requires Node.js 18 or newer.

2. Log in

savine login

Choose email at the prompt and enter your Savine credentials, or paste an API key if you already have one. The CLI writes credentials to ~/.savine/config.json (mode 0600) and verifies the session by pinging /api/v1/auth/me.

Confirm:

$ savine whoami
  Sarthak Gupta
  email : you@example.com
  id    : 2c28d2ad-…
  plan  : free
  auth  : api-key

3. Scaffold a project

savine init my-first-agent --template agent
cd my-first-agent

You now have:

my-first-agent/
  config.yaml          # declarative manifest — name, limits, tools
  agent.py             # entrypoint — exports run(input: str) -> dict
  requirements.txt     # python deps
  README.md

Open agent.py and customise the run() function — or leave it untouched for your first deploy.

4. Deploy

savine deploy

▲ savine  deploying agent from config.yaml

  bundling /Users/you/my-first-agent
  uploaded 4 files: config.yaml, agent.py, requirements.txt, README.md
✓ Agent deployed: my-first-agent
  id       fdc63413-…
  version  1
  status   READY
  url      https://savine.in/agents/my-first-agent
  linked to .savine.json

The .savine.json file now remembers this project, so you don't need to pass the agent id on later commands.

5. Run it

savine run --input "Say hi from my first Savine agent" --follow

--follow tails execution events as the task runs. The CLI streams THINK, TOOL_CALL, TOOL_RESULT, and OBSERVE events in real time and returns when the task finishes.

6. Iterate

Edit agent.py or config.yaml, then re-deploy. Every deploy creates a new immutable version.

savine deploy
savine agents versions            # list versions
savine agents info <id>           # full details

If a deploy regresses behaviour, roll back:

savine agents versions
savine agents rollback <id> <v>   # via systems/workflows, same shape

7. Observe

savine tasks ls -n 10             # recent runs
savine logs --follow              # live logs for the linked project
savine metrics overview           # usage + cost summary
savine open                       # jump to the dashboard

Common next steps

• Add a real model provider key: savine providers add openai --key sk-…
• Wire a graph of agents: System Spec
• Go programmatic: Python SDK
• Put this in CI: see the CLI Reference → CI/CD section
• Full command catalogue: CLI Reference