feat: initial Skeen-CRM AI Agent architecture

- FastAPI + Python 3.12 backend
- Meta WhatsApp Business API client (official)
- OpenAI GPT-4o with function calling
- RAG vector store with pgvector
- ERPNext Frappe REST client
- Celery + Redis async task queue
- PostgreSQL with migrations (Alembic)
- Docker Compose full stack
- Enterprise logging, metrics, health checks

alembic/versions/20260428_init.py

@@ -0,0 +1,100 @@
+"""Initial migration: conversations, messages, knowledge_chunks.
+
+Revision ID: 001
+Revises:
+Create Date: 2026-04-28 00:00:00.000000
+
+"""
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+from sqlalchemy.dialects import postgresql
+
+# revision identifiers, used by Alembic.
+revision: str = "001"
+down_revision: Union[str, None] = None
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    # Enable pgvector extension
+    op.execute("CREATE EXTENSION IF NOT EXISTS vector")
+
+    # Conversations table
+    op.create_table(
+        "conversations",
+        sa.Column("id", sa.String(36), primary_key=True),
+        sa.Column("phone_number", sa.String(20), nullable=False, index=True),
+        sa.Column("patient_id", sa.String(100), nullable=True, index=True),
+        sa.Column("patient_name", sa.String(255), nullable=True),
+        sa.Column(
+            "status",
+            sa.Enum("active", "paused", "resolved", "escalated", "appointment_confirmed", name="conversationstatus"),
+            nullable=False,
+            server_default="active",
+        ),
+        sa.Column("context", postgresql.JSONB(astext_type=sa.Text()), server_default="{}"),
+        sa.Column("last_message_at", sa.DateTime(timezone=True), nullable=True),
+        sa.Column("created_at", sa.DateTime(timezone=True), server_default=sa.func.now()),
+        sa.Column("updated_at", sa.DateTime(timezone=True), server_default=sa.func.now(), onupdate=sa.func.now()),
+    )
+
+    # Messages table
+    op.create_table(
+        "messages",
+        sa.Column("id", sa.String(36), primary_key=True),
+        sa.Column("conversation_id", sa.String(36), nullable=False, index=True),
+        sa.Column("direction", sa.String(10), nullable=False),
+        sa.Column("role", sa.String(20), nullable=False),
+        sa.Column("message_type", sa.String(50), server_default="text"),
+        sa.Column("content", sa.Text(), nullable=False),
+        sa.Column("whatsapp_message_id", sa.String(100), nullable=True),
+        sa.Column("tool_calls", postgresql.JSONB(astext_type=sa.Text()), nullable=True),
+        sa.Column("tool_results", postgresql.JSONB(astext_type=sa.Text()), nullable=True),
+        sa.Column("tokens_used", sa.Integer(), server_default="0"),
+        sa.Column("metadata", postgresql.JSONB(astext_type=sa.Text()), server_default="{}"),
+        sa.Column("created_at", sa.DateTime(timezone=True), server_default=sa.func.now()),
+    )
+
+    # Knowledge chunks table (for RAG)
+    op.create_table(
+        "knowledge_chunks",
+        sa.Column("id", sa.String(36), primary_key=True, server_default=sa.text("gen_random_uuid()::text")),
+        sa.Column("content", sa.Text(), nullable=False),
+        sa.Column("metadata", postgresql.JSONB(astext_type=sa.Text()), server_default="{}"),
+        sa.Column("category", sa.String(50), server_default="general"),
+        sa.Column("source", sa.String(255), server_default=""),
+        sa.Column("embedding", sa.String(), nullable=True),  # Stored as string; pgvector uses special type
+        sa.Column("created_at", sa.DateTime(timezone=True), server_default=sa.func.now()),
+        sa.Column("updated_at", sa.DateTime(timezone=True), server_default=sa.func.now()),
+    )
+
+    # Create pgvector column properly using raw SQL
+    op.execute("""
+        ALTER TABLE knowledge_chunks
+        ALTER COLUMN embedding TYPE vector(1536)
+        USING embedding::vector(1536)
+    """)
+
+    # Indexes
+    op.create_index("idx_knowledge_category", "knowledge_chunks", ["category"])
+    op.create_index("idx_knowledge_source", "knowledge_chunks", ["source"])
+    op.execute("""
+        CREATE INDEX idx_knowledge_embedding
+        ON knowledge_chunks
+        USING ivfflat (embedding vector_cosine_ops)
+        WITH (lists = 100)
+    """)
+
+
+def downgrade() -> None:
+    op.drop_index("idx_knowledge_embedding", table_name="knowledge_chunks")
+    op.drop_index("idx_knowledge_source", table_name="knowledge_chunks")
+    op.drop_index("idx_knowledge_category", table_name="knowledge_chunks")
+    op.drop_table("knowledge_chunks")
+    op.drop_table("messages")
+    op.drop_table("conversations")
+    op.execute("DROP TYPE IF EXISTS conversationstatus")
+    op.execute("DROP EXTENSION IF EXISTS vector")