openmaic.example.yml

# =============================================================================
# openmaic.yml: the model configuration of an OpenMAIC deployment
#
# Copy this file to openmaic.yml in the directory the server starts in (the
# repository root; /app in the Docker image), or point OPENMAIC_CONFIG at it,
# then restart the server. The file is read once at startup.
#
# As shipped, it needs one key: set OPENAI_API_KEY in .env.local (or change
# the provider below to one you have a key for). The commented blocks are
# optional; uncomment the ones you want and set the variables they name.
#
# - providers: the accounts the server can call.
# - slots: which provider and model each use of AI gets by default. Users
#   may change them in the model settings; slots left out are theirs to set.
# - lock: the slots fixed for everyone, each with its whole subtree.
# - allowUserKeys: whether users may add providers and keys of their own.
#
# Secrets stay in the environment (.env.local, a Docker or Kubernetes secret)
# and are referenced as ${VAR}. An unset variable, an unknown preset or slot,
# or a reference to an undeclared provider stops the server at startup with
# the exact location of the problem.
#
# Full reference: packages/docs/content/docs/configuration.mdx
# =============================================================================

providers:
  # A provider id (the key) is lowercase letters, digits and dashes. Slots
  # refer to models as "<provider id>:<model id>".
  openai:
    preset: openai
    apiKey: ${OPENAI_API_KEY}

  # --- Optional providers ----------------------------------------------------

  # Another chat vendor.
  # anthropic:
  #   preset: anthropic
  #   apiKey: ${ANTHROPIC_API_KEY}

  # Web search.
  # tavily:
  #   preset: tavily
  #   apiKey: ${TAVILY_API_KEY}

  # Most presets cover one capability: OpenAI images are a second provider
  # with the openai-image preset (the same key works).
  # openai-image:
  #   preset: openai-image
  #   apiKey: ${OPENAI_API_KEY}

  # A token plan covers several capabilities with one key: chat, images,
  # video, text to speech and web search for MiniMax.
  # minimax:
  #   preset: minimax
  #   apiKey: ${MINIMAX_API_KEY}

  # Any OpenAI-compatible server: baseUrl is required, and models lists what
  # the web settings offer from it.
  # local:
  #   preset: openai-compatible
  #   baseUrl: http://host.docker.internal:11434/v1
  #   models: [qwen3:8b]

  # Services that authenticate with a key pair use credentials.
  # alidocmind:
  #   preset: alidocmind
  #   credentials:
  #     accessKeyId: ${ALIDOCMIND_ACCESS_KEY_ID}
  #     accessKeySecret: ${ALIDOCMIND_ACCESS_KEY_SECRET}

slots:
  # The default chat model. Every chat slot without an assignment of its own
  # follows it.
  llm: openai:gpt-5.5

  # --- Optional slots (each needs its provider above) -------------------------

  # An object adds call parameters and a fallback model, tried once when the
  # primary fails with a retryable error after its own retries. Fallbacks
  # apply to language-model slots (llm and the slots below it) only.
  # course.outline:
  #   model: anthropic:claude-sonnet-4-6
  #   fallback: openai:gpt-5.4-mini

  # The Pro workbench agent needs a model with tool calling. It may set
  # thinking.mode but not thinking.effort; an effort set on llm is dropped
  # for the agent.
  # agent:
  #   model: openai:gpt-5.5
  #   api: openai-completions

  # A provider id alone means that provider's default model (search and
  # document services mostly have none to pick).
  # webSearch: tavily
  # image: openai-image:gpt-image-2
  # tts: minimax:speech-2.8-turbo

  # null turns a capability off by default; users may still turn it on
  # unless lock names it.
  # video: null

# Slots fixed for everyone, with every slot below them ("lock: all" fixes
# every slot). Each must be written under slots above (null counts).
# lock: [video]

# false: users cannot add providers, keys or token plans; they only choose
# among the providers declared above, for the slots lock leaves open.
# allowUserKeys: true