Our production monorepo (private): 4,739 commits, 76 branches, shipped and maintained daily by the OpenMetrix team.
AllMetrix is an AI-powered business management platform delivered through Telegram. Every employee in an organization — founder, manager, or specialist — works with a personal AI agent that plans their day, tracks tasks and business metrics, coordinates the team, and remembers everything the organization has ever discussed.
The platform is closed-source. This repository accompanies our President Tech Award
application as a technical overview; deeper source access for the evaluation committee
can be arranged on request. See NOTICE.md for terms.
- Role-aware AI agents — each employee gets an agent matched to their role (Founder / Manager / Specialist), with permissions and capabilities to match
- Daily operating rhythm — automated morning briefings and evening reviews, delivered in each employee's own timezone and working hours
- Tasks, reminders, and team coordination — assignment, overdue tracking, cross-team reporting, and proactive nudges
- Business metrics — employees log metrics conversationally; founders get plans-vs-facts analytics and statistics reports
- Long-term organizational memory — the AI recalls decisions, facts, and context from the organization's entire history
- External integrations — Google Sheets, CRMs, and other business tools connected directly by the agent through our integration hub
- Voice and media — voice messages (speech-to-text and spoken replies), photos, documents, and video understood natively
- Multi-language — English, Russian, and Uzbek
- Multi-tenant by design — every query and scheduled job is strictly scoped to the organization
- Python 3.13 · FastAPI · SQLAlchemy (async) · Alembic · Pydantic
- LangGraph / LangChain agent runtime with a multi-provider LLM strategy (Google Gemini, Anthropic Claude, OpenAI)
- PostgreSQL · Redis · Qdrant (vector search) · MongoDB
- BGE-M3 self-hosted embedding & reranking inference service
- Celery for distributed scheduling and background jobs
- python-telegram-bot client surface; Whisper STT and TTS for voice
- Docker Compose deployment
Production issues in an AI product are invisible without deep instrumentation, so observability is a first-class subsystem of the platform:
- Langfuse & LangSmith — end-to-end tracing of every agent turn, LLM call, and tool execution, with prompt/response capture for debugging and quality review
- Sentry — real-time error monitoring across the API, bot, and background workers
- Prometheus — service health and business metrics exported from every component
- Structured logging (Loguru) — key-value logs correlated across services
- LLM cost accounting — per-provider token and cost tracking on every model call, feeding weekly cost and usage reports
- Scheduler heartbeats — liveness monitoring of the background job dispatcher, so silent scheduling failures page us instead of the customer noticing
- 4,700+ commits of continuous delivery in a single production monorepo
- Pull-request workflow with automated AI code review on every PR
- Typed, linted, formatted codebase (
mypy,ruff,black) with a production test suite covering multi-tenancy isolation and retrieval quality - Full observability: structured logging, per-call LLM tracing, and cost accounting
© OpenMetrix Inc. All rights reserved. See NOTICE.md.