I'm a solo founder building a productised AI operating system — not an agency, not the next Morningside or Klarus. One vertical first (UK self-storage), then rinse-and-repeat. Niche within a niche, on purpose.
The end goal is a closed loop: the AI wired into everything the business runs on — Slack, meetings, email, docs, their core platform — so nobody hand-feeds it context. It understands the why behind decisions and acts unprompted. My bet: without that loop, you're just selling a chatbot, not solving real pain.
Architecture so far:
- A cloneable "bridge" that on setup generates per-engine config (CLAUDE.md / AGENTS.md / GEMINI.md) + an Obsidian vault for the human view. Engine-agnostic, no data, no keys.
- Separate "brain" repos = the data. One repo = one key = one hard permission wall, so personal can't leak into company.
- "Modules" = bolt-on capabilities. First one's a voice agent doing inbound/outbound/email.
- Git-backed, so it syncs to any device.
Sharing this openly on purpose — there's enough money in this space for everyone, and honest feedback is worth far more to me than the idea is to a copycat.
What I actually want from people who've built and sold this:
- The loop: how do you do ambient ingestion (Slack/meetings/email → distilled memory) without drowning in noise and cost? Git/markdown, DB, or vector store?
- Repeatable template: clone-and-run-a-script, or a real provisioning system? What does one-person-deployable look like?
- Onboarding non-technical clients without a support nightmare.
- Multi-tenant: updating/monitoring/versioning many clients centrally, no hand-touching.
- Pricing: real setup + monthly, SMB vs mid-market vs enterprise.
- Enterprise table-stakes: governance, security, data residency.
- Does a one-click version of this already exist that I've missed?
Roast it — where does this break?