24/7 customer support
A chatbot on your website or in your app, answering over your knowledge base. It recognises when it's unsure and hands the conversation to a human. Measurably reduces routine load on first-line support.
// Service · AI agents for businesses
AI agents for small and medium businesses.
Custom AI assistants built for your processes. They answer on your website, on WhatsApp, Messenger, Telegram and SMS. We work over your data, connect to your e-shop and business systems, and never lock you in to a single model vendor.
// About the service
Software that can independently work with your data.
By AI agent we mean software that can independently perform tasks over your data — answer questions, work with documents, call your APIs, and escalate to a human when needed. It's not just a chatbot pasted onto a website. We build solutions that solve a concrete business problem: save people time, give customers faster answers, or pre-filter incoming leads.
// Messaging agents
An agent where your customers already write.
Customers don't want to hunt for a form on your website. They message you on WhatsApp or Messenger, or send a text. The agent replies right away, even on a Sunday evening, and takes the answer from your systems: order status from the e-shop, stock levels, a free slot in the calendar. Whatever it can't handle goes to a person, together with the whole conversation.
- WhatsAppThrough Meta's official WhatsApp Business Platform. No unofficial libraries that end with a banned number.
- MessengerThe agent handles messages to your Facebook page and routes the harder ones to your inbox.
- TelegramThe quickest to launch. Works well inside the company too: field technicians, warehouse, team alerts.
- SMSFor customers who don't bother with apps, and for appointment confirmations. Two-way, through an SMS gateway.
We connect it to your e-shop, ERP, CRM or accounting system. Via API, and where there is no API, directly through the database or a scheduled export.
Hi, when will order 240817 arrive?
Hi, order 240817 was handed to the carrier today at 11:40. Expect delivery tomorrow between 12:00 and 16:00. Shall I send you the tracking link?
Yes please. Could you add one more pack?
I can't add anything to a parcel that has shipped. I'm passing you to our sales team, they'll reply here within 15 minutes.
// What it's good for
Six typical scenarios.
Instead of a generic chatbot, we build agents focused on one specific problem. Below are typical scenarios where deployment makes sense for small and medium businesses.
RAG over company documents
An agent that searches your internal documents, contracts or wiki and answers employees precisely, with a link to the source. Useful for large, fragmented knowledge bases where manual search would take minutes.
Email automation
Categorisation of incoming messages, draft replies, and summarisation of long threads. Always with human approval before sending — the agent prepares, the person approves. Saves hours daily without losing control.
Lead qualification from forms
The agent reads the inquiry, enriches it with public data about the company, scores its relevance and assigns it to a specific salesperson. Cuts out noise and saves hours of manual triage every week.
Internal CRM/ERP assistant
Natural-language queries over company data: "How many invoices did we issue to client X this year?" or "Show me projects running late." No need to teach people SQL or the database schema.
Structured meeting notes
From a recording transcript, the agent produces structured minutes, action items and calendar follow-ups. Especially useful where note-taking capacity is limited and important points get lost.
// How we work
We start from the problem, not the technology.
Standard flow from first meeting to production.
01
Free consultation
A one-hour session where we walk through your process and identify where AI makes sense (and where it would be overkill). Out of this comes a proposal and scope estimate.
02
Prototype on your data
We build a proof-of-concept on your real data. You get to try how it would work before you commit to anything. No obligation to continue.
03
Production deployment
Integration into your systems, monitoring, rate limits, fallbacks for LLM API outages. We also cover security: prompt injection, data leakage, audit logging.
04
Operations and iterative tuning
Models change, your data changes too. We regularly evaluate accuracy, add examples to RAG indexes and refactor prompts so the agent stays reliable.
// What we work with
Vendor-agnostic stack.
The model that's optimal today may not be optimal in a year. The architecture accounts for migration.
// LLM models
- Anthropic Claude (Opus 4.7)
- OpenAI GPT-5
- Google Gemini 2.5
- Mistral
- Llama 3 / 4 (on-prem)
// Frameworks
- LangChain
- LangGraph
- LangSmith
- MCP (Model Context Protocol)
- Claude Agent SDK
- LlamaIndex
- Pydantic AI
- Vercel AI SDK
// Vector DBs
- Pinecone
- Qdrant
- Weaviate
- Chroma
- pgvector
// Backend
- Python
- TypeScript
- Go
// Infrastructure
- AWS Bedrock
- Google Vertex AI
- Modal
- Together AI
- self-hosted / on-prem
// Security
- Prompt sanitisation
- Output filtering
- Per-user rate limits
- Audit logging
// FAQ
What SMB clients usually ask.
How much does AI agent development cost?
Pricing is always agreed up-front as a fixed scope. A prototype typically takes 2–3 weeks of work, production deployment 6–12 weeks depending on integration complexity.
If you don't have an idea yet, we'll give you a realistic scope on the introductory call based on your scenario.
How quickly will we see something working?
First working prototype within 3 weeks. The point of the prototype is so that you can test on your own data whether AI makes sense in the specific scenario before spending more.
Do we own the source code?
Yes. All code is yours. We host either on your infrastructure or in the cloud under your account. No vendor lock-in from our side.
What about our data — GDPR, sensitive information?
Data stays with you. When using external LLMs (Anthropic, OpenAI), we use zero-retention APIs and contractual Data Processing Agreements.
For the most sensitive data we deploy local models (Llama 3, Mistral) on your infrastructure — data then never leaves the company.
What if OpenAI or Anthropic raise prices or shut down their API?
The architecture is vendor-agnostic. Migrating between models is a configuration change, not a code rewrite.
If needed, we can switch from a cloud LLM to on-prem within a week.
Can you work entirely on-prem, without internet?
Yes. Local models (Llama 3 70B, Mistral Large) run on your hardware. They require 1–2 GPU servers, but data stays 100% inside your company.
What do we need to run the agent on WhatsApp?
A business account with Meta and a phone number for WhatsApp Business. We handle business verification and message template approval with you.
When a customer writes first, the agent can reply freely for 24 hours. Outside that window a business may only send an approved template, such as an order confirmation or an appointment reminder. Meta charges for templates; regular replies to customers are free.
Let's talk if AI agents make sense for you.
A free hour-long call. We'll figure out together if it's worth pursuing — and if so, in what scope.
Book a consultation →
// Contact
Let's work together.
Describe your situation or request and we typically reply within one business day. Free, no obligation.
- E-mailinfo@hruby.software→
- Phone+420 211 221 865→
- StudioPod Slovany 1888/12, Prague 2→
- HoursMon–Fri · 09:00–18:00 CET→