AI Solutions & System Integration
Boutique web studio building the API layer and MCP protocol that connect AI to business data, with help managing API keys, budgets and quality monitoring to technical standards.
Bringing AI (ChatGPT, Claude, Gemini) into a company's operations is more than dropping in a ready-made chat window or installing a packaged automation plugin. Leaning on packaged tools often brings dependence on third-party infrastructure, leaks of internal data and API costs that are hard to control.
Start with the operating architecture
Perxel approaches AI solutions as a technical partner, focused on building a dedicated API layer and applying the MCP protocol (Model Context Protocol). Whether it builds on an existing platform such as WordPress or on a standalone Next.js application, Perxel works on a pay-for-what-you-use basis, so the business keeps its budget in check and full ownership of its digital assets.
Three pillars of AI work at Perxel
- A flexible architecture. Perxel puts chatbots and AI directly on existing data (WordPress, Next.js) when the structure is clean enough. When needed, it builds a standalone system and syncs data both ways to keep operations efficient.
- Cost that follows real use. Connect through an AI gateway such as OpenRouter or call APIs directly, with a measuring tool such as Langfuse to keep answer quality and budget transparent.
- Ownership and governance. The business holds the API key account. Perxel helps set budget limits and hands over clean code, with no recurring licence fee.
What gets built
API layer and MCP layer
An intermediate API layer and the MCP protocol let AI models query company data safely in real time. This model already runs in production through the OpenWallet project, which gives the Owie AI assistant and external AI tools structured bank card data through a dedicated MCP server.
Chatbots on company data
WordPress is a typical data source for a chatbot built directly on top, when the structure is clean enough. For more complex systems, Perxel builds the chatbot on Next.js or as a standalone system with two-way data sync, to keep it safe and fast.
Automating large-scale data processing
AI applied to operating problems with a large amount of data. On the Khatra Tile project, Perxel built the Perxel AI Translate plugin to translate close to 1,000 posts and products into English, with pay-for-what-you-use API costs far below fixed-fee services.
AI governed to technical standards
1. Monitoring output quality
Every query is traced through Langfuse, answer accuracy is evaluated, replies that fall short are flagged and prompts are tuned continuously.
2. Managing API keys and budget
The business holds the API key account. Perxel helps set spending limits (budget caps) and automatic alerts so running costs stay within reach.
3. Pay-for-what-you-use cost
Connect through a familiar AI gateway such as OpenRouter, or work directly with the APIs of the model providers (OpenAI, Anthropic, Google), billed on the tokens actually used.
When it fits
- When a business wants a chatbot or AI connected directly to its existing operating data.
- When a standard API or MCP layer is needed so an AI assistant can query data safely.
- When the volume of content to process (translation, summarising) is large enough that API automation costs less than a fixed-fee service.
- When the API budget needs close monitoring and control instead of being left to a packaged provider.
FAQ
Contact Perxel
Footnotes
Want a quote?
Get in touch with Perxel today.