[ PRIVATE AI ]
An AI that knows
your business.
And keeps it private.
Most AI asks you to trade your data for intelligence. Actaa refuses the trade. You get an AI grounded in everything your company knows, and everything it knows stays yours.
[ NO FINE PRINT ]
What "private" actually
means here.
Not a marketing word. It is an architecture: four commitments, enforced in the product itself.
- 01
Your data never trains a model
All model and embedding processing runs on AWS Bedrock, which under AWS’s terms does not use your content to train foundation models. Actaa operates no training or fine-tuning pipeline of its own.
- 02
No consumer AI services
Every language-model and embedding call runs on AWS Bedrock inside our own AWS account. Nothing is sent to a consumer AI product or a model vendor’s own API. Speech-to-text runs on an open-source model on our own servers.
- 03
Isolated per company
Postgres row-level security on every table that carries a company id, forced so even the application’s own database role cannot read past it, with a second check in the application layer. Tables that hang off a parent record are isolated in the application layer.
- 04
Personal data is redacted
Redacted when your content is indexed, again on the question itself, and a third time on the model’s answer before it reaches you. HR-sensitive content is blocked from the AI path entirely.
Each of these is written down, not just claimed. See How Actaa AI Works, the AI Data Processing Addendum, and the AI Acceptable Use Policy.
[ WHERE WE ARE ]
Private is a ladder.
We’ll always tell you which rung.
“Private AI” means four different things depending on who is selling it. Here is exactly what it means here, today, and where it goes next.
- 01
Today · Enterprise-isolated — live
Enterprise model infrastructure inside our own AWS account. No training on your data, nothing sent to a consumer AI service, and isolation enforced per company in the database itself.
- 02
Next · Dedicated inference
Reserved model capacity for your organisation rather than shared throughput. On the roadmap, not shipped.
- 03
Then · On-device
Inference that never leaves hardware you control. Where this is going.
Most of this category calls rung one “private” and stops talking. We’d rather tell you it’s rung one.
[ GROUNDED, WITH RECEIPTS ]
Answers from
your reality.
With sources.
Ask it anything about the business. Every answer is drawn from your company’s real files, scoped to what you are allowed to see.
"What's our runway after the Q3 hires?"
11.4 months, assuming both offers out are signed and Q3 revenue lands at forecast.
[ THE INTELLIGENCE FLOOR ]
Six ways in.
One brain.
Intelligence is a whole floor of the product, straight from the sidebar.
01
Conversation
Ask anything about the business. Answers grounded in your data, with sources.
02
Roundtable
One question, a panel of models. Compare answers before you decide.
03
Agents
Delegate a defined piece of work to a specialist agent, run it on demand, and see every tool call it made, governed by your workspace AI budget.
04
Data Room
Bring a spreadsheet into a session and ask it questions. Every answer is computed, not guessed, and the raw data self-deletes in 72 hours.
05
MCP Conversation
Use Actaa's intelligence from Claude Desktop or your IDE, on the org's terms.
06
History
Every conversation kept and searchable. Institutional memory, not lost tabs.
And the same brain lives inside Write, Tables, Present and Meeting notes, drafting, modeling and summarizing where the work happens.
[ THE ENGINE ]
Every model.
One private engine.
Actaa runs on AWS Bedrock inside our own security boundary, enterprise model infrastructure trusted by banks, with smart routing that picks the right model per query in Conversation, and retries on an alternate model when one fails. And it is governed like a system of record.
Smart routing
The best model per task, with automatic fallback.
Budgets & controls
Per-user AI budgets your admins set and see.
Audit trails
Every Conversation request logged with model, tokens, latency and any block reason. Admins can review their workspace’s AI chats in Settings.
[ AND YOUR IP ]
Protected like it's
worth protecting.
Because it is. Beyond privacy, Actaa actively defends what you make.
- 01
Sealed in an IP ledger
The documents you create and the files you upload get tamper-evident proof of what is yours, and when.
- 02
Copies get caught
If your work surfaces in another company, it is flagged.
- 03
Share on your terms
Watermarked previews and no-download tiers for sensitive files.
[ ONE CONVERSATION ]