Client pilots underway in Sydney · Early-adopter places open
A self-hosted, access-controlled knowledge layer for your firm. It ingests your clients, projects and SOPs, stays fresh on a schedule, and is queryable by your staff's own AI tools — with per-person access scoping. Built on GBrain, the open-source shared-brain stack from Y Combinator's Garry Tan.
Two problems land at once for businesses getting serious about AI.
Your knowledge is siloed — locked in email threads, SharePoint folders and individual heads — so every person and every AI session starts cold. Ask ChatGPT about your client and it knows nothing. Ask the new hire and they ask the person next to them.
And most products that fix the silo do it by locking your institutional knowledge inside their platform. Solve the silo, inherit the lock-in — and in a field where the best model, agent and workspace change quarterly, lock-in is the expensive kind of mistake.
A Company Brain fixes the silo without inheriting the lock-in.
GBrain — git-backed, plain-file knowledge — running in your cloud account or ours, with access locked to your enrolled devices and tools.
Every item in a source is sorted ingest / redact / exclude, and you sign off before anything goes in. Regulated or high-risk data is excluded, not worked around.
Email, SharePoint/OneDrive, or Slack on a scheduled refresh — the brain stays current within 24 hours, not a stale export from last quarter.
Internal and HR material isolated from client material, enforced at the database layer and demonstrated live — not promised by convention.
Each teammate queries the brain from the AI tools they already use — Claude, ChatGPT, Cursor — over MCP, with their own credentials and their own scope.
“What's the latest on client X?” answered with citations on your real data — plus an architecture one-pager, access register and restore procedure.
Answers cite their sources, and where the brain doesn't know, it says so. Freshness within 24 hours on the connected source. Access scoping verified live with no cross-boundary leak. The wow-flow demonstrated on your real data at handover.
Your knowledge lives as plain, git-backed files in your own environment, on an open-source (MIT) stack. That means you keep freedom of movement on everything else: swap AI models, swap agent harnesses, swap chat platforms — the brain stays. If we part ways, you keep everything, in open formats.
This isn't theoretical. We've already swapped our own brain stack once — it cost us an afternoon, not our knowledge. That's the architecture working.
One thing we're upfront about: your staff's AI tools send approved, scoped excerpts to commercial LLM providers under enterprise no-training terms. Owning your knowledge means you can take it anywhere — not that it never touches a model.
Client pilots with Sydney businesses are underway now, and we're taking a small number of additional early-adopter clients. The pilot wedge is deliberately low-risk: one team, one source, one working wow-flow on your real data — proven before anything expands.
Early adopters get founding-client attention and direct access to the consultant who runs the stack — we use the same architecture on our own business, every day. We publish what we learn as the pilots progress, and nothing goes live in your business without the same classification, access-scoping and oversight discipline we apply to everything else.
GBrain is an independent open-source project — we're not affiliated with or endorsed by its creators. We build on it because it's the best stack for the job.
A Company Brain is a self-hosted, access-controlled knowledge layer for your business. It ingests your clients, projects, SOPs and communications, stays fresh on a schedule, and is queryable by your staff's own AI tools (Claude, ChatGPT, Cursor) over MCP with per-person access scoping. We build Company Brains on the open-source GBrain stack and are running client pilots with Sydney businesses now.
Chatswood Computer Consultants is a Sydney-based consultancy running GBrain — the open-source shared-brain stack from Y Combinator's Garry Tan — in production, on our own business and in client pilots with Australian businesses. We handle the full build: data classification, self-hosted deployment, connector setup, per-user access scoping, and staff MCP connections.
No — that's the design principle. A Company Brain stores your knowledge as plain, git-backed files in your own environment, on an open-source (MIT) stack. You can swap AI models, agent harnesses or chat platforms at any time and keep the brain; if we part ways, you keep everything in open formats. To be clear about what this does and doesn't mean: your staff's AI tools still send scoped queries to commercial LLM providers under enterprise no-training terms — sovereignty here is ownership and portability of your knowledge, not an air gap.
Nothing is ingested before a data-classification audit: every item in a source is sorted ingest / redact / exclude, and you sign off on the classification. Regulated or high-risk data (for example clinical or payroll) is excluded, not worked around. At query time, your staff's AI tools send approved, scoped excerpts to commercial LLM providers (Anthropic, OpenAI) under enterprise no-training terms. We're upfront about that — no air-gap pretence. If offshore processing is a blocker, the same models can run in-region via AWS in Sydney, so your data stays onshore. And because the brain is model-agnostic, a fully self-hosted model is possible for strict-privacy environments — with real cost and capability trade-offs we'll walk you through honestly.
It's a focused, fixed-scope build measured in days, not months, once access and classification sign-off are in place: server build, first source connected on scheduled refresh, per-person access scoping tested live, staff connected over MCP, and a working demonstration on your real data at handover.
Pilot commercials are agreed per engagement while the program is in its pilot phase. Published pricing lands when the program exits pilot, expected late 2026. Early-adopter clients get founding-client terms and direct access to the consultant who runs the stack.
A working demo on your own data, off a tight pilot. That's the ask — and the offer.