HUXBERRY
Infrastructure Review

G-Brain Installation Review

A review of the gbrain knowledge-brain deployment, how it's being used, how to wire it into your AI tools, and how (and whether) it ingests email, CRM, and NetSuite.

Prepared for  Nish Faria Reviewed  1 June 2026 System  gbrain v0.41.26.0 @ brain.huxapps.com Status  Reviewed · fixes applied same day

Update — 1 June 2026 · actioned the same day

Since this review the gaps below were fixed and verified live in the brain:

  • Sync gaps closed: suppliers 6 → 71 pages (64+ stranded local pages committed & pushed), products 196 → 197, systems-handbook merge conflict resolved (18 → 19). Brain now holds 481 real wiki pages.
  • Clients rewired: Codex now reaches gbrain via mcp-remote; the stale Codex→Hermes MCP entries were repointed from the decommissioned Hostinger host to vps2 (handshake verified).
  • Skills retargeted: /query now hits gbrain semantic search first; /ingest now enforces commit + push so authored pages actually reach the brain.
  • Remaining (you): one-time Claude Code /mcp → gbrain → authenticate browser step.
Executive summary (as reviewed)

The brain was built and healthy — but almost nothing read from it

Net effect at review time: you'd built a brain, then kept reading the filing cabinet. The retrieval-quality win (hybrid keyword + vector + knowledge-graph) was sitting unused on your interactive tools. None of it was broken architecture — and most of it is now fixed.

01 — Installation

What's deployed, and how healthy it is

gbrain v0.41.26.0 runs in Docker on vps2 (Hetzner) as gbrain-http + gbrain-postgres (pgvector), fronted by brain.huxapps.com through a Cloudflare Tunnel. The /admin path sits behind Cloudflare Access + Google SSO; /mcp behind OAuth 2.1. Its internal enrichment LLM runs through a loopback Codex bridge (your ChatGPT subscription), so marginal cost is roughly $1–5/month (embeddings only).

Locked architectural principle

The wikis are the system of record; gbrain is a search projection, never the source of truth. Everything flows wikis → git → brain, never the reverse. Keep this in mind for every recommendation below.

Live health (vps2, after same-day fixes)

481
Real wiki pages
904
Embeddings
310
KG links
0
Integrations configured

Total page count is 843 — that includes 362 pages in the inert legacy default corpus (pre-v0.17, never synced, harmless). The 481 figure is the real, searchable wiki content across 9 sources (up from 413 at review time, after the suppliers fix).

Sources

SourcePagesLast sync
products1972026-06-01
company-context952026-05-29
suppliers712026-06-01fixed — was 6
vps-architecture412026-06-01
odoo-implementation222026-05-28
systems-handbook192026-06-01conflict resolved
lead-gen172026-05-28
training102026-05-28
assets92026-05-28
default (legacy)362neverinert

Red flags — found at review, now resolved

✓ Fixed this session

  • suppliers had 6 pages, not ~68: root-caused to 64+ pages (materials/, vendors/, intelligence/) authored locally but never pushed to GitHub. Committed, pushed, re-synced → 71 in the brain.
  • products had 10 locally-modified pages uncommitted (stale in brain) — committed & pushed.
  • systems-handbook had an unresolved git stash pop conflict in index.md — resolved (kept all three additive rows) + processes.md added.
  • Codex pointed at a missing wrapper (the old gbrain serve pattern is obsolete on the v0.41 thin-client) — rewired to mcp-remote.

Remaining

Claude Code (Mac) still needs the one-time /mcp → gbrain → authenticate browser step. The hourly wikis-sync → gbrain-sync timers are confirmed healthy — the gaps above were all authoring-side (unpushed edits), not sync failures.

02 — Usage

How it's been used (and why "30 days" doesn't apply yet)

The brain was installed 2026-05-28 — days before this review, so there's no 30-day history. In its first few days:

03 — Making your AI use it

How to point Claude & Codex at gbrain

Two layers: fix the plumbing, then retarget the skills. Both were actioned this session.

Layer A — wire the clients

Layer B — retarget the skills (done)

wiki-query now calls mcp__gbrain__query with source_id:"__all__" (cross-wiki — gbrain does not federate by default) first, files only as fallback. wiki-ingest now ends with a mandatory commit + push step — the exact gap that left 64 suppliers pages invisible to the brain.

04 — Auto-ingestion

Email, CRM & NetSuite: what actually flows in

Today, the brain's only live intake is the wiki Git repos, pulled hourly and synced into Postgres. Nothing else flows in automatically.

The indirect path that already exists

Your Hermes agents (huxberrypm Gmail intake, Meet ingestion, scrapers, assetops) write markdown into a wiki's raw/inbox/, which becomes brain content on the next sync. So email and meeting data reach the brain only if a Hermes job or you ingest it into a wiki first. This keeps the "wikis = source of truth" principle intact.

gbrain's native "senses" (available, none enabled)

SenseSourceStatus
email-to-brainGmail (via ClawVisor gateway or Google OAuth)available · not set up
calendar-to-brainGoogle Calendaravailable · not set up
meeting-syncCircleback transcriptsavailable · not set up
x-to-brainTwitter / Xavailable · not set up
NetSuite / Odoo (CRM)no native sense exists

Recommended approach for CRM & NetSuite

NetSuite is lower priority since it's migrating to Odoo. The native email/calendar/meeting senses are an alternative for raw intake, but they bypass wiki review — only worth it if you specifically want a searchable firehose.

Status of recommended next steps

Where each follow-up landed

  1. ✓ DoneMac clients: Codex rewired to mcp-remote; Hermes MCP entries repointed to vps2. (Claude Code browser auth remains your one manual step.)
  2. ✓ DoneSkills retargeted: /query uses gbrain first; /ingest enforces commit + push.
  3. ✓ DoneSync health: root-caused to unpushed local edits; suppliers / products / systems-handbook all fixed, pushed, and verified in the brain.
  4. OpenEmail / CRM / NetSuite ingestion: plan agreed (Hermes→wiki curated summaries). Drafting the Odoo CRM-summary job is the next concrete step when you want it.