Shared Package

Why

Agent and Server were duplicating db.py (MongoDB connections + collection references) and schema.py (document builders like call_history_schema, conversation_history_schema). Changing a collection name or adding a field meant updating two places easy to forget, guaranteed bugs.

Now there's one source of truth: shared/.

What's in it

shared/
├── __init__.py
├── pyproject.toml
├── db.py          # All MongoDB collections
└── schema.py      # All document schema builders (DBSchema)

What's NOT in it

• Analysis uses async motor driver (not pymongo), has its own domain-specific schemas. No duplication.
• client_backend Node.js/Mongoose, different language entirely. Reads Python-written collections with strict: false.

How to use

Import from shared instead of local src.db / src.schema:

# Before
from src.db import call_history_collection
from src.schema import DBSchema

# After
from shared.db import call_history_collection
from shared.schema import DBSchema

Local development

Install the shared package in editable mode from Agent or Server directory:

pip install -e ../shared

Or add to requirements.txt:

-e ../shared

Docker

Build context is now the repo root (not individual service folders). The Dockerfiles handle it:

# Copies shared into the image
COPY shared/ /app/shared/
RUN pip install --no-cache-dir /app/shared

# Then copies the service
COPY Agent/ .

docker-compose.dev.yml uses context: . with dockerfile: Agent/Dockerfile (or Server/Dockerfile).

Production docker-compose.yml uses pre-built images shared code is already baked in at build time. No runtime dependency between containers.

Scaling

Shared code is copied into each Docker image at build time. At runtime, Agent and Server are fully independent containers. Scaling one doesn't affect the other.

When you change shared/

Rebuild both images:

docker compose -f docker-compose.dev.yml build agent server