Rockfort Shield · early deployment
Clients trust you with their data. None of it should reach the model.
Shield sits between your product and the model. Sensitive fields are masked before the model sees them, and restored on the way back.
Early deployment. We take a small number of design partners at a time.
How it works
Masked on the way in. Restored on the way out
Point your LLM API calls at Shield. Your application code and your prompts stay as they are.
Summarise the claim for Priya Raman, card 4111 1111 1111 1111, diagnosis E11.9.
Summarise the claim for [NAME_1], card [CARD_1], diagnosis [HEALTH_1].
Priya Raman's claim covers two outpatient visits. The card on file ends 1111.
Illustrative example
- Added latency
- Under 10ms for most requests
- Where it runs
- Inside your VPC. Your data stays in your environment
- Integration
- The same API. No rebuild, no prompt changes
- Logs
- Every interception, kept in your infrastructure
Coverage
What Shield catches
Sensitive fields
Works across
- Detection rules are configurable, including your own entity types.
- One integration covers your AI apps, agents and copilots.
What early deployment means, said plainly
- Shield is running with a small number of design partners.
- You work directly with the team building it.
- We tell you what it catches and what it still misses, before you rely on it.
- Your feedback shapes what ships next.
Questions we get
Will this slow down our API calls?
Shield adds under 10ms to most requests. For most applications that is imperceptible, and we can share benchmarks for your stack.
We use more than one model provider. Does Shield cover all of them?
Yes. OpenAI, Anthropic, Gemini, Azure OpenAI and self-hosted models, through one integration.
Can we control what gets detected?
Yes. Define custom entity types, set thresholds, and decide how each data type is handled. The defaults cover the common cases from day one.
Where does our data go during processing?
Nowhere. Shield runs inside your VPC, and Rockfort does not store, process or see your application data.
Become a Shield design partner.
We take a small number at a time. Start with a 20-minute call to see whether Shield fits your stack.