Why Theresa
Most coding agents start with a ticket. Theresa starts with the problem.
Coding agents are powerful when you already know what to build. Theresa is built for the messy production questions your team has not scoped yet:
Add retry to the payment webhook
Acceptance criteria attached
They start where the thinking has already been done, from a ticket someone else scoped.
- Why are dashboards wrong?
- Why did infra spend spike?
- Why is ranking quality falling?
- Why is the ML pipeline degrading?
She starts where the thinking hasn't happened yet, from symptoms nobody has turned into a ticket.
Built for complex live systems
The places where the code is already running, the symptoms are real, and nobody has written the ticket yet.
Recommendation and ranking systems
Recall coverage, ranking quality, diversity, and cold start.
Data pipeline reliability
Silent drops, late-arriving events, schema drift, stale features.
Cloud and ML infra cost reduction
Where the spend actually goes, and what it buys you.
Dashboard and metric correctness
Numbers that disagree, and the joins and definitions behind them.
Production monitoring and repair
What broke, why it broke, and a fix that holds.
Upgrading your legacy SaaS with an AI-native customer system
The workflows your customers actually touch, rebuilt AI-native on top of the SaaS you already run.
Measured on real engagements
How It Works
Diagnose. Prioritize. Fix. Remember.
You decide what we can see.
See exactly where time and money leak.
Ship whatever the problem needs.
Every engagement makes the next one sharper.
CAPTURED FROM EVERY ENGAGEMENT
Your data stays yours.
We observe only the work you approve, and nothing else. No keystrokes. No productivity scores. Sensitive fields are redacted on your machine, before anything leaves it.
What kind of data does AI Theresa collect?+
AI Theresa only observes approved work activity, not personal identifiers. The desktop client does not record keystrokes or mouse movements, and is not designed to score individual-employee productivity. Sensitive fields are filtered client-side before any data leaves the device, with field-level redaction available.
Is my company's data isolated from other customers?+
Yes. Each customer's data, workflow memory, and shipped automations remain in org-level isolation. Nothing is shared across customers.
Is data encrypted? Where are credentials stored?+
Data is encrypted in transit and at rest. OAuth tokens and credentials live in a hardened secret-management layer.
Can we export or delete our data?+
Yes. Customers can request export or deletion at any time.
Do you use my data to train your models?+
Customer data is not used to train shared or third-party models. Workflow memory only informs that customer's own automations.
Who controls who can see the workflow intelligence?+
Customer-controlled. You decide who in your organization can see aggregated workflow intelligence.
How long do you keep our observation data, and what happens when we cancel?+
Raw observation events are kept on a 90-day rolling window; derived insights are kept until you delete them. On termination, full deletion completes within 30 days; certification of deletion is available on request.
Does AI Theresa take screenshots of user computers?+
No. By default, AI Theresa does not capture screenshots of user computers. Screenshot capture can be enabled opt-in for specific approved workflows when a customer requests it.
Give your team a senior production architect on demand.
Tell us where it hurts. We'll come back with what AI Theresa would find inside your company.
Technology partners
OpenAI
Anthropic
Google Cloud
Microsoft
AWS