Configuration

Configuration is loaded from YAML through EngineConfiguration.

Related pages: Engine, EngineConfiguration-Reference, EngineConfiguration-Services-Reference, Overview, Troubleshooting.

File resolution rules

When no inline config is provided:

1. Read ENV from process environment.
2. If absent, try to bootstrap ENV from dotenv secret adapter in config.yaml.
3. If ENV exists and config.<ENV>.yaml exists, use it.
4. Otherwise use config.yaml.

If neither exists, engine startup fails.

Selection behavior is implemented in EngineConfiguration.load_configuration().

Secret templating in YAML

YAML is rendered with Jinja2 and can reference secrets:

services:
  vector_store:
    vector_store_adapter:
      adapter: "qdrant"
      config:
        host: "{{ secret.QDRANT_HOST }}"
        api_key: "{{ secret.QDRANT_API_KEY }}"

Two-pass behavior:

• First pass loads only secret service to resolve secret placeholders.
• Second pass renders full configuration with resolved secrets.

If a secret is missing and no TTY is available, loading fails with an explicit error.

Secret lookup precedence during rendering:

1. Process environment variable.
2. Bootstrap dotenv adapter (first pass only).
3. Configured secret service adapters (second pass).
4. Interactive prompt (TTY only, second pass).

For full behavior details, see EngineConfiguration-Reference.

Top-level sections

• api: title, description, logos, CORS, reload.
• global_config: ai_mode in cloud | local | airgap.
• services: adapter selection and adapter-specific config.
• modules: enabled module entries and module config payload.
• deploy (optional): deployment metadata.
• default_agent (optional): runtime default agent string.

Built-in defaults

If service blocks are omitted, defaults are applied:

• object_storage: filesystem (storage/datastore)
• triple_store: filesystem (storage/triplestore, triples)
• vector_store: qdrant_in_memory
• secret: dotenv (.env)
• bus: python_queue
• kv: python

Default module behavior

If not explicitly configured, naas_abi_core.modules.templatablesparqlquery is auto-added.

This happens via model validation in EngineConfiguration.validate_modules().

Minimal complete example

See Quickstart for a ready-to-run local config.yaml.

Advanced reference

For class-by-class documentation of EngineConfiguration.py, see EngineConfiguration-Reference.

For adapter-by-adapter service and deploy configuration fields, see EngineConfiguration-Services-Reference.