# gbrain: emerging patterns, and who it leaves out (FINAL — Sam's edit)

Gbrain, Garry Tan's open-sourced knowledge base, which he runs his own agents on, is at almost 20k Github stars and has processed 146,000 pages across all its users.

Here are a couple of patterns from analyzing the repo that show us where company brains and knowledge bases are heading:

-> Reasoning on top of retrieval. It's not enough to just retrieve and cite information, you need to be able to synthesize and make judgements on top (e.g. if there is contradicting information, which piece do you trust?)

-> Dynamic lens on the data. The schema for Gbrain's knowledge graph isn't fixed, you can load a new one in based on your preferences. As humans, we're able to view the same set of information from multiple perspectives. So should our agents. A strict, predefined hierarchy is very quickly outdated, no matter how MECE it is.

-> Don't reinvent isolation. Per-person access rides on existing infrastructure. The database solved multi-tenancy decades ago; Reuse it instead of building a new one.

-> Copy data in, don't live-query it. Meetings, emails, tweets, voice get ingested from their sources and reasoned over as one stable corpus, not refetched from ten APIs on every question.

-> A brain needs verbs. Gbrain comes with 43 skills which can do things like trace the trajectory for decision, search for conflicting information, or brainstorm over a folder. These are what get people to actually use it.

But Gbrain isn't built to be a mass market product. Here's the catch. It is built by a technical person, for technical people. You install it by pointing an AI agent at a protocol, or bun install and a Postgres, or an MCP config with OAuth.

The people who'd get the most out of a company brain are exactly the ones that don't fit that profile. They're the ones who don't know how to use GitHub to do version control or navigate complex database schemas.

Our open-sourced knowledge base is built around that premise. Instead of feeling like a developer tool, it's meant to feel like Notion, except the main users are your agents. As more knowledge work is done by AI, this will become increasingly what the winners in the category will focus on.

Drop me a DM or message if you're curious about learning how a company brain can automate your work, reduce on-boarding time, and align your team!