01In production
Agent platform
Twelve services in production. Eight agent types — support, collections, appointments, after-sales and more — across WhatsApp, Telegram and webchat. Behaviour control, traceability and isolation by design.
Designing solutions with artificial intelligence
The best answer for your case is not always a conversational agent.
We design and build the whole solution: the architecture, how the data is handled, privacy, where it runs, what gets measured and which model is worth tuning. And when agents are the way, we do not start from scratch — Agentry, our agent management platform, is already in production.
Real autonomy inside a defined path
How we work on your case
We start from the problem, not from the tool. These are the disciplines we bring to the table; which ones apply, and in what order, depends on what we find.
Our own product
Agentry
Our agent management platform.
When the solution involves agents, the work starts on a platform already running in production, not on a blank page.
See what it includes- 01
Solution architecture
We define what to automate, what to leave to a person and how it fits what already works. Sometimes the answer is an agent; sometimes it is a shorter process and a dashboard.
architecture · integration
- 02
Data analysis and metrics
We look at what your operation already produces — conversations, tickets, transactions — to find where time is lost, and we leave it measured: what resolves on its own, what is handed over and what each conversation costs.
exploration · cost per case · resolution
- 03
Privacy, compliance and cloud
Masking of personal data, isolation per customer and access logs. Where it runs is your call, and that decision shapes the design from day one — it is not something bolted on at the end.
PII · residency · on-premise · air-gapped
- 04
Agent development
Agents that complete a request end to end, connected to your systems, with behaviour limited to what the organisation authorised.
FSM · tools · channels
- 05
Tuned models
When the case justifies it, we tune a model on your data or deploy an open one in your infrastructure. When it does not, we say that too.
fine-tuning · open models · evaluation
Agentry, the foundation
The product much of the work stands on. Four layers inside one perimeter: the agent platform is in production; the next ones extend models, teams and infrastructure without stepping outside that control.
- 02In development
Client-tuned models
Multi-provider and open models running on client infrastructure, chosen agent by agent according to what each task needs.
- 03In development
Trained teams
Training so your organization can operate, configure and evolve agents without depending on us. The panel already allows it; the formal programme is on the way.
- 04Coming soon
Infrastructure and hardware
On-premise and air-gapped deployment already available. Certified hardware as an appliance is the next step — when we can sustain it operationally.
How it looks in your industry
Four real cases. None is an FAQ chatbot: each completes an end-to-end process.
Early collections without friction
- The problem
- The call centre dials thousands of clients with minor arrears. Calling is expensive and annoys those who would pay anyway.
- What the agent does
- The agent writes on WhatsApp, verifies identity, shows the real balance and offers only authorized options. Disputes go to an agent with context.
Early collections no longer depends on call centre size.
Appointments without phone calls
- The problem
- Reception handles the phone all day for appointments while people at the desk wait.
- What the agent does
- Finds real availability, books, reschedules and confirms. Frees the slot if the patient cannot attend.
Less no-shows and reception focused on who is present.
After-sales that does not get lost
- The problem
- "Where is my order?" dominates messages. Answered manually, office hours only.
- What the agent does
- Checks real shipment status, handles returns per your policy and generates labels when due.
Customers resolve alone, any time.
Help desk for your team
- The problem
- The same access and policy questions every day, answers scattered in documents nobody finds.
- What the agent does
- Answers from internal manuals, creates tickets and escalates to the right team.
Less waiting and more time for what matters.
What your security team will ask
Straight answers, for the first meeting with procurement and risk.
Where does my data live?
Wherever you decide: our cloud to start in days, a dedicated cloud in the country you need with a residency contract, or your own server — even with no internet connection.
cloud · dedicated · on-premise
Does my data train a model?
No. We do not train models on your conversations. The AI runs with your key or in your infrastructure, and our price does not go up with your volume.
your AI, your account
Can data mix between customers?
No. The database prevents it at the architecture level, customer by customer. Documents, cards and phone numbers are stored masked.
isolation · masking
Who can see the conversations?
Only who you allow, with per-role permissions and an immutable log. Every incoming message is HMAC-verified and credentials are encrypted.
roles · HMAC · encryption
Can I take my data with me?
Yes. Conversations and traces are exportable at any time. Nothing is held hostage.
exportable
Who is on the other side
Nobody signs with an anonymous vendor. Here is what you need to know.
We are Dendra. We have built enterprise software for years, and on every artificial intelligence project we hit the same wall: the business wanted to automate a complete request, and risk asked what would happen if the model promised something the company could not deliver. Nobody had a good answer.
So we turned the problem around: instead of asking the model to behave, we built Agentry, where going off-script is impossible. But Agentry is the foundation, not the whole catalogue — before proposing an agent we look at the whole problem, and if the shortest path is another one, we say so.
- Operations
- GlobalOperations
- Site languages
- 6Site languages
- Deployment available
- On-premDeployment available
For your systems team
What they will look for when you forward this page.
Restricted behaviour, in action
Greeted and introduced itself as the bank assistant
Example: a collections case, step by step
Why the model cannot leave the flow
Each state declares valid transitions; model output is validated against that set before it is applied.
FSM · 4 state types · type validation
What is recorded
Event per state entry/exit, model call and tool invocation, with prompt, latency, tokens and cost.
trace events · prompt_version · cost_usd
Per-client isolation
RLS in Postgres, tenant prefix in Redis, own collection in Qdrant. tenant_id resolved once at the gateway.
Postgres RLS · Redis · Qdrant
Connection to your systems
HTTP with allowlist, read-only SQL, scripts in a restricted environment. Encrypted credentials.
http · db_query · AES-256
Supported models
Per agent: commercial provider with client key or open model on their infrastructure.
OpenAI · Anthropic · Gemini · local
Observability
Metrics, logs and traces on an open stack, integrable with your current observability.
OpenTelemetry · Prometheus · Loki
Frequently asked questions
What if my case does not need an agent?
It happens often. Many problems are solved by better data handling, a missing integration or a dashboard nobody built. We say so in the assessment, even when it means selling less.
How is this different from a button chatbot?
A button tree only answers what was foreseen. This understands what the customer writes, queries your information and completes the process. The difference shows in the edge case.
How long does it take, and what do you need from my team?
A first agent on a bounded case is usually a matter of weeks; what stretches the timeline is access to internal systems and security review. After that, the day to day — configuration and copy — is handled from a panel: your systems team gets involved once, at the start.
What about cases the agent cannot resolve?
They go to your team with the full conversation, in a prioritized queue. The customer stays in the same chat, repeating nothing.
How is it priced?
Solution design is billed per project, with a written and bounded scope. Agentry is an annual subscription per agent in production, plus support level. AI usage you pay to the provider: our price does not go up with your volume.
Let us evaluate your case
Thirty minutes. Bring the process that takes the most time from your team.