Backoffice framework-native migration notes

This package reduces direct legacy dependencies in the active runtime.

Main changes:

Domain nodes call framework services through src.compat.framework_services.
IMDB enrichment uses MCP tool consultar_imdb_cliente first.
Speech enrichment uses MCP tool consultar_speech_analytics first.
TAIS KB enrichment uses MCP tool consultar_tais_kb first.
Direct TIM clients are disabled by default and only available behind BACKOFFICE_ALLOW_LEGACY_CLIENTS=true for parity tests.
Legacy model aliases such as bo_gptoss20b_dev are mapped to the framework model, defaulting to openai.gpt-4.1.
Non-serializable progress producers are no longer stored inside LangGraph state, reducing checkpoint hash corruption risk.
MCP server defaults to mock mode for local development.

Recommended local execution:

# Terminal 1: MCP mock server
uvicorn mcp_servers.backoffice_mcp_server.main:app --host 0.0.0.0 --port 8010 --reload

# Terminal 2: backend
uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload

Optional standalone IMDB mock, only if you intentionally enable legacy direct clients:

uvicorn mock_servers.mock_imdb_server:app --host 0.0.0.0 --port 8011 --reload

For a fully framework-native local test, keep:

BACKOFFICE_ALLOW_LEGACY_CLIENTS=false
BACKOFFICE_MCP_USE_MOCK=true
BACKOFFICE_MCP_BACKEND_TYPE=mock
CLASSIFICATION_LLM_MODEL=openai.gpt-4.1
CLASSIFICATION_LARGE_LLM_MODEL=openai.gpt-4.1
TAIS_KB_LLM_MODEL=openai.gpt-4.1

1.5 KiB Raw Blame History

Backoffice framework-native migration notes

1.5 KiB

Raw Blame History