One-command deployment for Cogitator agents — Docker and Fly.io targets with auto-detection.

Overview

The @cogitator-ai/deploy package builds, packages, and deploys your agent application. It auto-detects your server adapter, required services, and LLM secrets from your project config.

pnpm add @cogitator-ai/deploy

CLI

# Deploy to Docker (default)
cogitator deploy

# Deploy to Fly.io
cogitator deploy --target fly

# Dry run — preview without executing
cogitator deploy --dry-run

# Check status
cogitator deploy status

# Tear down
cogitator deploy destroy

Programmatic API

import { Deployer } from '@cogitator-ai/deploy';

const deployer = new Deployer();

// Preview — run preflight checks only
const plan = await deployer.plan({
  projectDir: process.cwd(),
  target: 'docker',
  noPush: true,
});

console.log(plan.preflight.checks);

// Deploy
const result = await deployer.deploy({
  projectDir: process.cwd(),
  target: 'fly',
});

console.log(result.url); // https://my-app.fly.dev

Configuration

Add a deploy section to cogitator.yml:

deploy:
  target: fly
  port: 3000
  region: iad
  image: my-agent
  registry: ghcr.io/myorg/myapp
  services:
    redis: true
    postgres: true
  secrets:
    - OPENAI_API_KEY
    - DATABASE_URL
  resources:
    memory: 512mb
    cpu: 1

Auto-Detection

The deploy engine reads your project to determine what it needs:

What	Source	Example
Server adapter	`package.json` dependencies	`@cogitator-ai/express` → Express
Services	`cogitator.yml` memory config	`adapter: redis` → Redis service
Required secrets	LLM provider in config	`openai` → `OPENAI_API_KEY`
Ollama Cloud	Model suffix	`:cloud` → `OLLAMA_API_KEY`

Deploy Targets

Docker

Generates a Dockerfile and docker-compose.prod.yml in .cogitator/:

cogitator deploy --target docker          # Build image only
cogitator deploy --target docker --push   # Build + push to registry

TypeScript projects get a multi-stage build (builder → runtime). JavaScript projects get a single-stage build.

Fly.io

Generates fly.toml and deploys via flyctl:

cogitator deploy --target fly --region lhr

Requires flyctl installed and authenticated (flyctl auth login).

Architecture

ProjectAnalyzer  →  ArtifactGenerator  →  DeployProvider  →  Result
(detect config)     (Dockerfile, etc.)     (docker/fly)       (url, status)

ProjectAnalyzer — reads package.json and cogitator.yml to detect server, services, secrets
ArtifactGenerator — generates Dockerfile, docker-compose, fly.toml from templates
DeployProvider — executes preflight checks, builds, deploys (Docker or Fly.io)
Deployer — orchestrator tying it all together

Custom Providers

import type { DeployProvider } from '@cogitator-ai/deploy';

class KubernetesProvider implements DeployProvider {
  readonly name = 'kubernetes';

  async preflight(config, projectDir) { /* ... */ }
  async generate(config, projectDir) { /* ... */ }
  async deploy(config, artifacts, projectDir) { /* ... */ }
  async status(config, projectDir) { /* ... */ }
  async destroy(config, projectDir) { /* ... */ }
}

const deployer = new Deployer();
deployer.registerProvider(new KubernetesProvider());

Deploy Package