Show HN: I built an open-source AI data layer that connects any LLM to any data (github.com)

Bag of Words is an open‑source “AI data layer” that lets teams connect any LLM to any data source with centralized context, governance, and observability. It provides a chat-driven interface (web, Slack, etc.) to query databases and services—Postgres, Snowflake, BigQuery, Redshift, dbt, Tableau and many more—and to generate charts, dashboards and scheduled reports. The system enriches prompts with contextual artifacts (dbt models, Tableau sources, code, AGENTS.md, KPI/rule definitions), runs an agentic loop for tool use, reasoning and reflection, and records every decision and trace so outputs can be audited and improved. Technically it’s model‑agnostic and deployment‑flexible: you can plug in OpenAI, Anthropic, Google Gemini, self‑hosted providers (vLLM, LM Studio, Ollama) or multiple providers at once, and swap models or data sources without breaking workflows. Enterprise features include RBAC, SSO/OIDC, audit logs, VPC deployment via Docker/Compose, VMs or Kubernetes, plus observability and quality metrics to track accuracy and feedback. For AI/ML teams this reduces vendor lock‑in, improves trustworthiness of LLM‑powered analytics, and provides a reproducible, governed pathway to build agentic, data‑aware applications and self‑improving pipelines.