# Quickstart Ten minutes, no paid model call. The CLI provisions a project and creates a safe nonproduction deployment revision; curl renders the prompt and sends a synthetic log. Every object can also be created in the web app at `https://app.prompton.ai`. You need `curl` and `jq` for the examples. Readable project, prompt, model and deployment references come from slugs, keys, provider/model strings, version numbers and revisions. Use a new UUIDv7 log ID for a new synthetic call, and keep the same ID only when retrying that call. ## 1. Install the CLI and sign in ```sh curl -fsSL https://raw.githubusercontent.com/polimo-dev/prompton-cli/v0.8.0/install.sh | PTN_VERSION=v0.8.0 sh prompton login --host https://app.prompton.ai ``` The CLI prints a browser link and a code: ```text Open this URL to approve the login: https://app.prompton.ai/device?code=H4KP-T7WR Your code: H4KP-T7WR Waiting for approval... ``` In the browser, sign in with the email code PromptOn sends you, then approve the CLI device. A new email address becomes an account on its first sign-in. The `--host` flag keeps the CLI pointed at the same PromptOn app as these docs. The host is remembered in `~/.config/prompton/config.json`. With one organization the CLI selects it; with several, run: ```sh prompton use --org personal ``` ## 2. Create a project ```sh prompton projects create helpdesk \ --description 'Customer support app' \ --idempotent --json > project.json PTN_SETUP_ENV=$(jq -r '[.environments[] | select(.slug != "production")][-1].slug // empty' project.json) test -n "$PTN_SETUP_ENV" || { echo "No writable nonproduction environment found"; exit 1; } prompton use --project helpdesk ``` New projects start with `production` and `staging`. Production is protected. New deployment revisions start in the lowest active nonproduction environment, which is `staging` for a default project. If `--idempotent` reused an existing project with more environments, `PTN_SETUP_ENV` follows that project's returned pipeline order. Promotion to production is a separate web or management API action. ## 3. Create a prompt One prompt represents one LLM call site in your app. The key is the stable integration name. ```sh cat > variables.json <<'EOF' [ { "name": "question", "type": "string", "required": true, "description": "The customer's message" }, { "name": "plan", "type": "string", "required": false, "description": "free or pro" } ] EOF prompton prompt create support_reply \ --name 'Support reply' \ --description 'Answers a customer support question' \ --variables-file variables.json \ --default-params '{"temperature":0.5}' \ --idempotent --json ``` The app can also discover variables for you when a prompt version or editor draft uses new Liquid variables. The variables file is still useful because it records descriptions, examples, and required fields. ## 4. Commit prompt content Create the managed Chat messages. Conversation history stays in your app; PromptOn stores the reusable managed messages, usually system instructions and prompt-specific context. ```sh cat > messages.json <<'EOF' [ { "role": "system", "content": "You are a friendly support agent for Acme. Answer in two or three sentences; if you are not sure, say so and offer to escalate." }, { "role": "user", "content": "{{ question }}" } ] EOF prompton prompt commit support_reply \ --file messages.json \ --message "migrated from the app's hardcoded prompt" \ --json ``` Versions are immutable. Committing a version changes nothing at runtime until a deployment pins it. ## 5. Register the model Register the provider/model pair you want to use. Deployments use the same readable pair; UUIDs are not accepted as resource selectors. ```sh prompton models register openai/gpt-4o-mini --idempotent --json > model.json ``` For OpenRouter models, the server fills display name, pricing, context length, and capabilities from the public catalog when it can. This is not a model inference call. ## 6. Deploy to the setup environment ```sh prompton deploy support_reply \ --environment "$PTN_SETUP_ENV" \ --provider openrouter \ --model openai/gpt-4o-mini \ --params '{"temperature":0.3}' \ --version latest \ --json > deployment.json DEPLOYMENT_REVISION=$(jq -r '.revision' deployment.json) PROMPT_VERSION_NUMBER=$(jq -r '.prompt_version_number' deployment.json) ``` The response is a deployment event: ```json { "revision": "v2026.09.03-1", "sequence": 1, "environment": "staging", "model": "openai/gpt-4o-mini", "provider": "openrouter", "prompt_version_number": 1, "params": {"temperature": 0.3}, "provider_options": {}, "created_at": "2026-09-03T04:13:40.204518Z" } ``` A revision is a pin: one prompt version number, one provider/model pair, params, and provider options. It is live for the setup environment as soon as the command succeeds. The example response assumes the default `staging` setup environment; if `PTN_SETUP_ENV` resolves to another environment, your response shows that slug. ## 7. Issue an SDK key ```sh PTN_KEY=$(prompton api-keys issue --name 'Helpdesk server' --quiet) ``` The secret is printed once. It belongs to the project, carries `read` and `logs` by default, and can read any environment in that project. ## 8. Render the deployed prompt This fills the template and returns the provider request shape. It does not call OpenAI, OpenRouter, or any other model provider. ```sh curl -sS "https://app.prompton.ai/api/v1/prompts/support_reply/render" \ -H "Authorization: Bearer $PTN_KEY" \ -H "content-type: application/json" \ -d "$(cat <