Omi Backend Patterns Skill

This skill provides guidance for working with the Omi backend, including conversation processing, memory extraction, chat system, and LangGraph integration.

When to Use

Use this skill when:

• Working on backend Python code in backend/

• Implementing new API endpoints

• Processing conversations or extracting memories

• Working with the LangGraph chat system

• Integrating with Firestore, Pinecone, or Redis

Key Patterns

Conversation Processing

The conversation processing pipeline follows this flow:

• Audio arrives via WebSocket (/v4/listen )

• Transcription via Deepgram/Soniox/Speechmatics

• Conversation creation in Firestore (status: "in_progress")

• Processing trigger via POST /v1/conversations or timeout

• LLM extraction of structured data:

• Title and overview

• Action items

• Calendar events

• Memories (user facts)

• Storage in Firestore and Pinecone

Key Function: utils/conversations/process_conversation.py::process_conversation()

Memory Extraction

Memories are extracted from conversations using LLM:

from utils.llm.conversation_processing import _extract_memories

memories = await _extract_memories( transcript=transcript, existing_memories=existing_memories, )

Categories: personal, health, work, relationships, preferences

Chat System Architecture

The chat system uses LangGraph for routing:

• Classification: requires_context() determines path

• Simple Path: Direct LLM response (no context needed)

• Agentic Path: Full tool access with LangGraph ReAct agent

• Persona Path: Persona app responses

Key File: utils/retrieval/graph.py

Module Hierarchy

CRITICAL: Always follow the import hierarchy:

• database/

• Data access (lowest)

• utils/

• Business logic

• routers/

• API endpoints

• main.py

• Application entry

Never import from higher levels in lower levels!

Database Patterns

• Firestore: Primary database for conversations, memories, users

• Pinecone: Vector embeddings for semantic search

• Redis: Caching (speech profiles, enabled apps, user names)

• GCS: Binary files (audio, photos, speech profiles)

API Endpoint Patterns

• Use FastAPI routers in routers/

• Keep routers thin - business logic in utils/

• Use dependency injection for auth

• Return consistent error formats

Common Tasks

Adding a New API Endpoint

• Create router function in appropriate routers/*.py

• Add business logic in utils/

• Use database functions from database/

• Follow error handling patterns

• Add to router in main.py

Processing Conversations

• Use process_conversation() from utils/conversations/process_conversation.py

• Handle extraction results

• Store in Firestore and Pinecone

• Trigger app webhooks if needed

Adding a Chat Tool

• Create tool function in utils/retrieval/tools/

• Use @tool decorator from LangChain

• Add to tool loading in utils/retrieval/tools/app_tools.py

• Tool will be available in agentic chat path

Related Documentation

The docs/ folder is the single source of truth for all user-facing documentation, deployed at docs.omi.me.

• Backend Deep Dive: docs/doc/developer/backend/backend_deepdive.mdx

• View online

• Chat System: docs/doc/developer/backend/chat_system.mdx

• View online

• Data Storage: docs/doc/developer/backend/StoringConversations.mdx

• View online

• Transcription: docs/doc/developer/backend/transcription.mdx

• View online

• Backend Setup: docs/doc/developer/backend/Backend_Setup.mdx

• View online

• Backend Architecture: .cursor/rules/backend-architecture.mdc

Related Cursor Resources

Rules

• .cursor/rules/backend-architecture.mdc

• System architecture and module hierarchy

• .cursor/rules/backend-api-patterns.mdc

• FastAPI router patterns

• .cursor/rules/backend-database-patterns.mdc

• Database storage patterns

• .cursor/rules/backend-llm-patterns.mdc

• LLM integration patterns

• .cursor/rules/backend-testing.mdc

• Testing patterns

• .cursor/rules/backend-imports.mdc

• Import rules

• .cursor/rules/memory-management.mdc

• Memory management

Subagents

• .cursor/agents/backend-api-developer/

• Uses this skill for API development

• .cursor/agents/backend-llm-engineer/

• Uses this skill for LLM integration

• .cursor/agents/backend-database-engineer/

• Uses this skill for database work

Commands

• /backend-setup

• Uses this skill for setup guidance

• /backend-test

• Uses this skill for testing patterns

• /backend-deploy

• Uses this skill for deployment patterns