Frequently asked questions

An AI agent is a system that uses a large language model (LLM) to plan and execute multi-step tasks autonomously. Unlike a chatbot that responds to prompts, an agent can call external tools (APIs, databases, web search), make decisions, and complete goals with minimal human input. Examples: a customer support agent that resolves tickets end-to-end, a research agent that browses the web and synthesizes findings, or a lead qualification agent that scores and routes inbound leads.

RAG is a technique that gives an LLM access to your specific knowledge at query time — your documents, databases, policies, product manuals — rather than relying on the model's general training data. A RAG system ingests your content, splits it into chunks, embeds those chunks as vectors, stores them in a vector database, and retrieves the most relevant pieces when a question is asked. The retrieved content is then included in the prompt so the LLM can answer accurately, with citations, from your actual data.

MCP (Model Context Protocol) is an open standard created by Anthropic that defines how AI models connect to external tools and data. An MCP server is a service you build that exposes your internal capabilities — search your database, create records, query your API — as callable tools that any MCP-compatible AI client can invoke. Claude Desktop, Claude Code, and custom AI agents can all use MCP servers. Think of it as a standardized plugin system for AI.

Fine-tuning trains the model itself on your data — baking knowledge into the model weights. RAG retrieves your data at inference time and provides it as context. RAG is better for frequently-updated knowledge, large document collections, and cases where you need citations or verifiable answers. Fine-tuning is better for consistent style, specific output formats, or domain-specific reasoning patterns. We typically recommend RAG first because it's faster to build, easier to update, and more auditable — fine-tuning when RAG alone isn't sufficient.

GPT-4o is a strong default — fast, multimodal, and excellent at instruction-following. Claude 3.5 Sonnet has the best performance on code generation and nuanced reasoning tasks, with a 200K context window useful for long documents. Gemini 1.5 Pro handles the longest contexts (1M tokens) and is competitive on multimodal tasks. For high-volume, cost-sensitive applications we often route simple queries to GPT-4o-mini or Claude Haiku and complex ones to the flagship models. We run benchmarks on your specific task during the discovery phase.

Book a free 30-minute strategy call using the contact form on our homepage. Tell us what you're trying to build, your timeline, and any relevant tech stack context. We'll come prepared with initial thoughts and questions. After the call, we'll send you a fixed-price proposal within 24 hours.

Yes. We sign mutual NDAs before discussing sensitive business details. For regulated industries (healthcare, legal, finance), we also sign Data Processing Agreements (DPAs). Our standard DPA is available at satekk.com/dpa.

Yes. We work with clients across the United States, UK, Europe, and worldwide. All communication is in English. We accommodate time zone differences for meetings and async communication via Slack works well across regions.

Yes, selectively. We work with agencies and consultancies that want to offer AI development services to their clients. If you have a client project that requires specialized AI engineering, reach out to hello@satekk.agency to discuss partnership terms.

Focused single-workflow automations or chatbots: $2,500–$5,000. Custom AI agents with tool use and integrations: $5,000–$15,000. Multi-agent systems: $10,000–$30,000+. SaaS MVPs: $8,000–$25,000. Marketing websites: $1,500–$8,000. Prices depend on scope, complexity, and integrations. Every project gets a fixed-price quote after a free discovery call.

We absorb the cost. We give fixed-price quotes, not time-and-materials estimates. If a build takes longer than we scoped, we don't charge more — we finish what we committed to. This is why we scope carefully before giving a price.

For project work: 50% upfront, 50% on delivery. For monthly retainers: monthly in advance. We accept bank transfer (ACH/wire) and major credit cards via Stripe.

Yes. We integrate AI systems into existing infrastructure — any database, CRM, ERP, helpdesk, or custom API. If it has an API or webhook support, we can connect to it. We've integrated with Salesforce, HubSpot, Intercom, Zendesk, Notion, Google Workspace, Stripe, QuickBooks, and dozens of custom APIs.

All data is encrypted in transit (TLS 1.3) and at rest. We follow the principle of least privilege — AI systems only access the data they need. For clients handling sensitive data (PII, health records, financial data), we build with additional isolation, audit logging, and compliance controls. We sign DPAs and can accommodate on-premise or private cloud deployments where cloud AI services aren't acceptable.

Yes. For clients who can't use cloud AI APIs due to data residency, compliance, or security requirements, we deploy with open-source models (Llama 3, Mistral) on your own infrastructure. Performance is slightly below frontier models but often sufficient for focused tasks — and the privacy guarantees are complete.

Yes. We build GDPR-compliant AI systems by design — data minimization, user consent flows, right-to-erasure implementation, DPA signing with all AI API providers (OpenAI, Anthropic process data under their own DPAs), and proper data retention policies. Our standard DPA is at satekk.com/dpa.