Reliable AI, connected to the way you already work
Your systems stay. We connect them, add AI that does the actual work, and deliver software your team really uses. Monitored and secured from day one.
AQUNAMA connects AI agents to the systems you already have. Nothing gets replaced, your data stays under your control in the EU or on your own servers, and the AI model can be swapped at any time.
Your systems
Nothing gets replaced.Integration layer
Moves data securely between systems.AI layer
Does the actual work.Outcome
What your team uses every day.
Architecture, integration and AI layers with clear boundaries: typed APIs and pipelines, RAG over governed data, agent orchestration with guardrails, and observability across the whole solution.
Your systems
Systems of record, unchangedIntegration layer
APIs · pipelines · access controlAI layer
Model-agnosticOutcome
Product · custom · hybrid
How we think about technology
We build exactly what we show you. Your stack stays. And you are never locked into our platform.
- We connect to what you have. Your systems stay, and we add AI to them.
- We choose the technology per project. Proven, mostly open-source components, with no vendor lock-in.
- Security by design. Security and compliance are part of the brief from the first day.
- Code is checked in several layers. Automated tests, AI code review and a security check. A senior engineer has the final word.
- Fixed scope, price and go-live date. In writing, before work starts.
Integration layer: we connect anything that holds data
AI is only as useful as the data it can reach. Your systems stay the systems of record, and AI works on top of them.
APIs and webhooks (REST)
Real-time connections to modern systems
Data and event pipelines (ETL)
Data moved between systems, with every run checked
Databases
For older and in-house systems, with only the access that is needed
Email and files
IMAP, Microsoft Graph, attachments, exports
Telephony (SIP, cloud PBX)
The agent answers calls, makes calls, transcribes and evaluates them
Model Context Protocol (MCP)
An open standard that lets our agents, and assistants such as Claude, ChatGPT or Copilot, use your data securely
Access is controlled by role (RBAC), and every transfer is recorded in an audit log. If an interface is missing, we build it.
AI layer: guardrails for live operation
Anyone can build an AI agent demo. An agent is ready for live operation only with these guardrails:
Agent orchestration
The agent acts only through defined tools (tool calling). Actions are idempotent, so nothing runs twice.
RAG over governed data
Answers come from your documents, and the agent sees only what it has permission to see. It cites the source or says it does not know.
Evals
Before launch and after every change, we test against the baseline on real, tricky and adversarial cases. Voice agents are tested with simulated calls.
Guardrails and policies
Prompt-injection protection, input and output checks, and limits for each client. Aligned with the OWASP Top 10 for LLM and agentic applications.
Human in the loop
A person approves sensitive actions, and the agent has only the permissions it needs.
Model-agnostic
OpenAI, Anthropic, Google and open-source models behind one LLM gateway. We switch models without rewriting anything.
Running AI: observability, cost and performance
Launch is only the start. We run AI as a critical system (LLMOps).
Observability
Tracing, metrics and logs for every model request and every phone call (OpenTelemetry). You see what the agent did and what it cost.
Cost optimization
A cheaper model wherever it is good enough (model routing), caching, and budgets and limits for each use case.
Performance
Low latency for voice, streamed responses and a fallback model during outages.
Safe changes
A new model or prompt first runs on a copy of live traffic (traffic mirroring). If there is a problem, we roll back to the previous version.
Security, data and deployment
We work with regulated industries such as insurance, finance and healthcare, including NIS2 requirements and internal audits.
Where it runs
Cloud (AWS, Azure, GCP)
In EU regions, with flexible scaling
Prague data center
Partner-operated, ISO 27001 and ISO 9001-certified, dedicated infrastructure in the EU
Your own servers (on-premise)
Including isolated (air-gapped) environments and GPU servers that run the models on site
Hybrid
Some parts in the cloud or the Prague data center, others on your infrastructure
Data and access
- Documented data flows, with each client’s data separated in the database itself (row-level security)
- AI providers chosen by region and data processing agreement (DPA), with no training on your data
- GDPR by design: retention, deletion, standard contractual clauses (SCCs) and a list of sub-processors
- Encryption in transit and at rest, SSO with MFA, and role-based access (RBAC)
- Supply chain security: only code that has passed tests, dependency checks and container scans goes to production
- Backups and recovery, with RPO and RTO agreed for each project
EU AI Act
At the start of every project, we classify each AI use case by risk category. The AI always introduces itself as AI, and important decisions about people stay with people.
Who owns what
Ownership and licensing are agreed in writing, up front.
Custom solutions
The full source code sits in your repository.
Our products (AI Voice Agent, AiLead)
We run them, so you start faster and at a lower cost.
A combination
With a clear line between what is yours and what is ours.
How a project runs
| Phase | What happens |
|---|---|
| Analysis | We find where AI pays off. Where it does not, we tell you. |
| Proposal | Scope, price, go-live date and the GDPR and AI Act assessment, in writing. |
| Build | You see results every week. |
| Launch | After validation against the baseline, usually about 6 weeks from signing. |
| Operation | Observability, cost control and quality tuning. |
Technical questions
Talk to an engineerDo we have to change our systems for AI?
No. We connect AI to what you have and build any missing interfaces.
Where will our data be, and who has access to it?
In the EU or on your own servers, as you decide. Only the roles that need access have it, and every access is recorded in the audit log.
Is it secure and compliant with GDPR and the AI Act?
Yes. We assess both at the start of every project, well before launch.
What if the AI makes a mistake?
The agent can take only approved actions, a person approves sensitive steps, and every step can be traced.
What if the model provider raises prices or shuts down?
We switch to another model without rewriting the application.
Who owns the code, and who maintains it?
We agree this in writing, up front. For custom solutions you get the full source code, and we either run the system ourselves or hand it over to your IT team.
Want to talk it through with our engineering team?
Tell us what you want to connect. We bring the people who will build it to the call.