Everyone's Suddenly Building Company Brains (9+ brains dissected)

Everyone’s Suddenly Building Company Brains

Source

LinkedIn post by Femke Plantinga (Aug 27, 2026) — 79 comments. Free interactive ebook with architecture notes from 149 teams: https://lnkd.in/eCU4BnyG

Everyone’s suddenly building company brains. Nobody agrees on what’s inside one. 😵‍💫

So they opened up 9+ company brains to see how they’re actually built. Every single one does the same four things: getting signals, remembering, dreaming & pruning, speaking & searching.

The 9 brains

1. GBrain — Garry Tan’s open-source personal brain. Email and calendar flow into a git repo; a nightly job re-links everything and flags what’s gone stale.
2. mem0 — A memory library you call from your own code. Only stores what you explicitly tell it; ranks fresh facts above idle ones at search time.
3. Letta — For building agents that remember across sessions. The agent decides what’s worth keeping; a second agent tidies up its memory in the background.
4. Zep / Graphiti — A knowledge graph with a clock in it. When a fact changes, the old one gets an end date instead of being overwritten — you can still ask what was true last March.
5. Sylph — A content brain that lives entirely in a git repo. Agents write drafts, humans publish, and afterwards the agent reads your edits to learn what it got wrong.
6. DIY (Claude Code + git) — What most engineering teams actually do. Markdown in the repo, grep instead of search, PRs as the only thing keeping it honest.
7. Plethor — A brand brain for marketing teams. Campaigns, assets and performance data in one tree; brand rules only move when a human signs off.
8. Gorgias Cortex — Built in-house by an 8-person AI team. 12,000 markdown nodes in GitHub; every night the questions it got wrong become PRs that fix it.
9. Slite Agent — For teams whose knowledge lives in docs and across sources. Watches ~20 connected tools (Slack, Drive, GitHub, Jira…) for staleness and sends the diff to whoever owns the page. Nothing changes without human approval.

Notable comments

• Fivos Aresti: “Gorgias turning wrong answers into PRs is a good loop as the brain gets better through corrections, not just more data.”
• JUNAID SAJID: “The ‘dreaming & pruning’ phase is often where most systems break down — stale or conflicting context can cause more issues than missing data. Pairing explicit memory (Letta/Mem0) with automated staleness checks (Slite Agent or PR-based curation) seems to be the sweet spot for production setups.”
• Yadullah Ali: “Retrieval is mostly a solved problem now. What kills these is that flagging a stale page is easy and nobody wants to own the page that got flagged.”

Key takeaway

The pattern across all 9: a memory system isn’t just storage — it needs a signal loop (what changed?), a pruning loop (what’s stale?), and a human ownership loop (who fixes it?). The ones that hold up all have a named human on the other end of the staleness signal.