Software your staff can trust, on a model you own.
Search with citations, agents with approval steps, and integrations that respect the permissions you already have.
What we build
Four kinds of system.
Search and answers
Answers drawn from your policies, contracts and records, each with a citation to the exact passage. Retrieval enforces your existing access rules, so users are only shown documents they’re entitled to; this is tested with permission-denied cases before release. The system says when it doesn’t know.
hybrid search · rerankers · permission-aware index
Search and answers: full details →Agents and workflows
Reviews, triage and drafting with the fewest permissions needed. By default, any action that writes data or moves money waits for a person to approve it, enforced in code, not in the prompt. Every action is logged.
tool allow-lists · approval steps · dry-run mode
Agents and workflows: full details →Integrations
Connections to clinical, banking and business systems through their standard interfaces, with single sign-on and your existing permissions.
HL7 FHIR · ISO 20022 · SAML / OIDC
Integrations: full details →Legacy → AI
We map and test the old system first, then change it in stages with tests that prove behaviour stays the same.
code mapping · characterisation tests · staged rollout
Legacy → AI: full details →What goes wrong
Common failures, and what we do about them.
| Failure | What we do |
|---|---|
| Hostile instructions hidden in a document or email | Treat retrieved content as data, not instructions, and test with planted attacks. |
| Users see documents they shouldn’t | Enforce permissions in the search index, not in the prompt. |
| Right document, wrong passage cited | Measure citation precision and highlight the exact span. |
| Tables split apart by chunking | Parse documents by structure before indexing. |
| Stale answers from an old index | Scheduled re-indexing with freshness checks. |
| Agents looping and running up cost | Step limits, budgets and alerts. |
| Quality drops after a model change | The full test suite runs before any swap. |
| Staff trust the output too much | Sources shown by default, and review steps where errors are costly. |
Frameworks
Oversight designed in.
| Framework | What it covers | How we design for it |
|---|---|---|
| OWASP LLM01, LLM06, LLM10 | Prompt injection, excessive agency, unbounded consumption. | Injection tests, least-privilege tools, step and cost limits. |
| EU AI Act Art. 14 and 50 | Human oversight for high-risk systems; telling people they’re dealing with AI. | Approval steps and clear disclosure in the interface. |
| MAS Guidelines on Artificial Intelligence Risk Management | Closer attention for AI that can act and use tools. | Action logs and approval records for every agent. |
Questions
What operations teams ask.
Can the assistant show a document someone isn’t entitled to?
No. Permissions are checked when documents are retrieved, using your existing access rules.
What can an agent do without approval?
Only read-only actions you list. Anything that writes or moves money waits for a person.
How do you stop a malicious PDF instructing the agent?
Retrieved content is treated as data, tools are limited, and we test with planted attacks before release.
What happens when the model is down?
The application fails safely and tells the user, and work falls back to the manual process.
What we don’t promise
Some claims we won’t make.
- Autonomous agents that move money or make clinical decisions.
- Clinical decision support without addressing medical-device rules first.
- Experience with a named system we haven’t integrated. We’ll tell you what we have built.
Put your model to work.
Start with one workflow, measured against how it runs today.