Build company brain for AI agents using graph context instead of plain RAG
As someone using AI agents for the last one year to run my company, I need them to understand company context, not just return related text chunks.
The problem: ask "what breaks if we deprecate the v1 API?" and standard RAG gives you four chunks from a design doc, a postmortem, a Slack thread, and meeting notes. The model has to still figure out on its own that the postmortem describes the same API the design doc deprecates, and that someone already posted a migration timeline in Slack.
I built a tutorial using HydraDB that adds graph context on top of vector retrieval. Instead of just ranked text, you also get relationship edges: billing-service DEPENDS_ON payments-api-v1, payments-api-v2 REPLACES payments-api-v1. Model gets structure, not a reading list.
The useful part was bring-your-own-graph. You declare service dependencies and team ownership explicitly instead of relying on LLM extraction. For structured data you already maintain, the graph is deterministic.
It also supports per-user memory. Same question, different depth depending on who's asking. An engineer gets migration mechanics. A manager gets timelines and ownership.
Runs end to end in 30 minutes with synthetic data.
Repo with full working code: https://github.com/manveer/company-brain-tutorial
Tutorial: https://hydradb.com/blog/build-company-brain-ai-agents