A private AI assistant is one that runs on infrastructure only your organisation uses — rather than a shared, multi-tenant service where your conversations, documents, and connected data sit alongside other customers' data.
The word "private" gets used loosely. For a lot of products it means "we don't train on your data" — which is a policy promise, not an architecture. For a genuinely private assistant, it means your data doesn't share a home with anyone else's in the first place.
The Three Kinds of "Private" You'll See
| Claim | What it actually means | Strength |
|---|---|---|
| "We don't train on your data" | A policy promise about how they use it | Weak — it's a promise, not a design |
| "Your data is encrypted" | Data is protected at rest and in transit | Good, but it's still on shared infrastructure |
| "Dedicated instance" | Your assistant and data run on infrastructure only you use | Strongest — isolation by architecture |
The first two can both be true of a shared service. Only the third changes where your data actually lives.
Why It Matters for a Business Assistant
This isn't abstract. Once an AI assistant connects to your tools — your CRM, your inbox, your client records, your internal docs — it's handling the same information you'd never paste into a public chatbot. The questions you'd ask it are the ones you'd never type into a free tool:
- "Summarise where the Johnson account stands."
- "What did we quote that client last quarter?"
- "Draft the renewal email using our actual contract terms."
Those answers only work if the assistant can see your real data. And that only makes sense if your real data isn't sitting in a shared bucket with strangers.
What to Check Before You Trust "Private"
- Is it shared or dedicated? Ask directly: "Is my instance shared with other customers?" The answer should be a clear no.
- Where does it run? Which cloud, which region — and can you find out for compliance purposes?
- How are credentials stored? Your connections to Slack, CRM, and email should be encrypted at rest, not held in plaintext.
- Can you get your data out? A private assistant you can't leave is a lock-in, not a service.
- Who can see the conversations? Your team — and nobody on the vendor side, unless you ask for support.
Privacy vs. Convenience Isn't a Trade-Off Anymore
The old assumption was that a genuinely private assistant meant self-hosting, a weekend of setup, and a model that was worse than the public ones. That's no longer the case. A managed private assistant gives you dedicated infrastructure without the ops burden — you get isolation by default, and the vendor runs the servers.
How Cody Handles Privacy
Cody is a managed AI assistant, and every customer gets a dedicated private instance — not a shared workspace. Slack tokens and integrations are encrypted at rest with AES-256-GCM. Setup takes about five minutes, and there's no lock-in: cancel anytime.
Because Cody is managed, "private" doesn't mean "you maintain it." We run the infrastructure; you get the isolation.
How Cody Compares
- Cody vs ChatGPT — a shared consumer service vs a dedicated private instance
- Cody vs Claude — shared infrastructure vs per-customer isolation
- Cody vs Dots (OpenAI) — a personal agent on OpenAI's infrastructure vs your own instance
- Cody vs Muse (Meta) — a consumer agent in Meta's cloud vs a private business instance
The bottom line: "we don't train on your data" is a promise. A dedicated instance is architecture. If your assistant is going to see your real business, check which one you're actually getting. See how Cody works →