# yaml-language-server: $schema=https://cubeship.dev/schema/template/v1.json
version: 1
name: 'Honcho'
# The first release whose instances create a managed Postgres with
# pgvector. Before it, the database would come up without the extension
# and Honcho's migrations would fail.
minCubeship: "0.9.0"
project: honcho

inputs:
  - key: domain
    type: domain
    label: Where the Honcho API answers
  - key: openaiApiKey
    type: secret
    label: Your OpenAI API key
    help: Honcho uses OpenAI for memory extraction, reasoning and embeddings. Model usage is billed by OpenAI.
  - key: jwtSecret
    type: secret
    label: The key Honcho signs access tokens with
    help: Keep this unchanged across updates. Generate a workspace access token after installing, as described in the README.
    generate: 48

databases:
  - key: db
    name: honcho-db
    engine: postgres
    version: "17"
    database: honcho
    # Honcho stores its embeddings as pgvector columns; its migrations
    # expect the extension to exist.
    extensions:
      - pgvector
    limits: { cpu: 1, memory: 1Gi }

  - key: redis
    name: honcho-redis
    engine: redis
    version: "7.4"
    limits: { cpu: 0.5, memory: 256Mi }

apps:
  - key: api
    name: honcho
    repo: https://github.com/cubeshipd/cubeship-templates
    ref: main:honcho
    build: dockerfile
    port: 8000
    health: /health
    domains:
      - host: ${input.domain}
    limits: { cpu: 1, memory: 2Gi }
    env:
      CUBESHIP_HONCHO_ROLE: api
      DB_CONNECTION_URI: postgresql+psycopg://${db.db.user}:${db.db.password}@${db.db.host}:${db.db.port}/${db.db.name}
      DB_POOL_SIZE: "5"
      DB_MAX_OVERFLOW: "5"
      CACHE_ENABLED: "true"
      CACHE_URL: redis://default:${db.redis.password}@${db.redis.host}:${db.redis.port}/0?suppress=true
      AUTH_USE_AUTH: "true"
      AUTH_JWT_SECRET: ${input.jwtSecret}
      LLM_OPENAI_API_KEY: ${input.openaiApiKey}
      VECTOR_STORE_TYPE: pgvector

  - key: deriver
    name: deriver
    repo: https://github.com/cubeshipd/cubeship-templates
    ref: main:honcho
    build: dockerfile
    # The deriver consumes the database queue; it serves no public HTTP API.
    limits: { cpu: 1, memory: 2Gi }
    env:
      CUBESHIP_HONCHO_ROLE: deriver
      CUBESHIP_HONCHO_API_URL: http://${app.api.internal}:${app.api.port}
      DB_CONNECTION_URI: postgresql+psycopg://${db.db.user}:${db.db.password}@${db.db.host}:${db.db.port}/${db.db.name}
      DB_POOL_SIZE: "5"
      DB_MAX_OVERFLOW: "5"
      CACHE_ENABLED: "true"
      CACHE_URL: redis://default:${db.redis.password}@${db.redis.host}:${db.redis.port}/0?suppress=true
      AUTH_USE_AUTH: "true"
      AUTH_JWT_SECRET: ${input.jwtSecret}
      LLM_OPENAI_API_KEY: ${input.openaiApiKey}
      VECTOR_STORE_TYPE: pgvector
      DERIVER_WORKERS: "1"