OpenAI, Claude, Vertex
Models
A fast model for classification, a stronger one for synthesis and logic.
Access through the API, keys, and a hard limit on token cost.
The cloud joined to a local Python script.
Agents and automation
AI implementation, autonomous agents, and custom business systems.
We turn the noise around artificial intelligence into saved time and higher revenue. We design, install, and join AI systems: from autonomous agents that take routine operational work, to a safe link between language models and your company's knowledge base.
- Codethe model is wired into the system, not a prompt on its own
- Agentsmulti-step work without constant supervision
- RAGan answer only from the company's documents
- Isolationclient data does not train a public model
A chat subscription is not yet an implementation.
- 1
A leak
Sensitive data lands in a public chat.
- 2
A hallucination
The model does not know the company, so it invents.
- 3
No connection
The CRM, the ERP, SQL, and the ads APIs stay outside the model.
Component
The model receives a chosen context, the result is checked, and the action lands in the system.
Scope
What we build.
PPC, competition, catalog
Agents
A PPC auditor reads hundreds of thousands of queries in Google Ads and Bing Ads and builds a negative package.
Monitoring reads changes in competitors' prices, offers, and ad structures.
Categorization matches supplier descriptions to Google categories and fills in the attributes.
RAG
Knowledge base
The model reads PDFs, sheets, Notion, or a CRM.
The answer cites only an internal source.
When the documentation is missing, the system says so outright.
Between systems
Processes
No more copying by hand between systems.
A document, a request for quote, and an email become a record in the ERP or the CRM.
A commercial proposal comes out of the brief.
Client data stays client data.
- 1
Commercial APIs
OpenAI, Anthropic, and Google Cloud do not train public models on this data.
- 2
Permissions
A read, or a human yes, before a deletion and before a payment.
- 3
Anonymization
Where it can be done, the data is cleaned before the model.
Questions
Does an AI implementation have to cost hundreds of thousands of PLN?
No. One process — reporting, lead classification, or a proposal assistant — starts from a few thousand to around fifteen thousand PLN. The return arrives within a few months.
Which technologies do you use?
Python and JavaScript or TypeScript, agent frameworks, vector databases, and the APIs of the current model generation: OpenAI (GPT-6 Astra, GPT-6.1 Sol), Anthropic (Claude Opus 5.5, Claude Sonnet 5.5), and Google (Gemini 4 Argon, Gemini 3.8 Flash). The cloud is Google Cloud or AWS.