AI Agent Development
AI agent development builds AI systems that go beyond fixed-rule automation — they hold conversations, make decisions, and run complex tasks autonomously. humaix designs agents for compliance risk assessment, lead generation, customer support, and internal briefing tools: systems that know your business, respect defined boundaries, and leave an audit trail, with LLM provider and hosting decided together with you.
In scope
- Agent architecture design: task definition, decision boundaries, human hand-off points
- Conversation/decision flow build: LLM integration, prompt engineering, tool definitions
- RAG/vector database connection and integration with internal data sources where needed
- Prompt-injection and edge-case testing, with documented boundary scenarios
- Audit trail and logging so every agent decision is traceable
- Integration with a monitoring dashboard or existing systems (CRM, help desk)
- Handover: documentation and baseline usage training for your team
Out of scope
- LLM provider selection and usage cost (decided together, billed separately)
- Server and hosting cost (billed at cost)
- Licensing and maintenance of existing software (ERP, CRM)
- Simple, single-step process automation (separate service: n8n automation)
Evidence
- P3Built a dialogue-based, RAG-backed compliance risk assessment agent for a German medical device manufacturer, with MDR and ISO 13485 checks designed into the architecture from the start.
- P2Indexed 21 medical textbooks and internal procedures into a vector database to power a support agent that cites its sources and hands off to a human when uncertain.
- Our production agent architectures run 225+ nodes — not demos, enterprise-scale systems handling thousands of operations a day.
- P1Built a WhatsApp sales agent that knows the business and its limits: message-type detection, prompt-injection testing, and documented edge cases.
Sectors
Pricing
| Scope | Duration | Range |
|---|---|---|
| End-to-end agent development | 6-12 weeks | 5,000 - 20,000 € |
| Monitoring and maintenance | — | 500 - 2,000 € |
- Prices exclude VAT.
- LLM usage and hosting are not included; they are billed at cost.
Frequently Asked Questions
How long does an AI agent project take?
A typical agent development project takes 6-12 weeks. The main drivers are how many systems the agent needs to reach, how complex its decisions are, and your regulatory requirements — testing takes longer in sectors like medical devices or finance.
Does the agent run on our own infrastructure, or does it use a cloud LLM?
Both are possible. The agent itself can run on your self-hosted infrastructure; on the LLM side you can choose a cloud provider (Claude, GPT) or a deployment scoped to your organization. We decide together based on your compliance requirements.
What happens if the agent makes a wrong decision?
Every agent is designed with a hand-off point to a human for uncertain or high-risk situations; we never leave an agent fully autonomous and unsupervised. Every decision is logged with an audit trail, so what happened and why stays traceable afterward.
Will it be GDPR-compliant?
Yes. Agents that touch personal data are built with data minimization, retention limits, and deletion flows from the start; with self-hosting, data never leaves your own infrastructure. We've built this architecture directly in regulated sectors like healthcare and finance.
Will it connect to our existing CRM, ERP, or help desk?
Yes, if the system exposes an API, we integrate directly. The agent is equipped with tools — writing to a CRM, reading from an ERP, opening a help-desk ticket — and we define upfront which tool has which permission.
How is this different from n8n automation?
n8n automation runs a predefined flow step by step; an agent makes its own decisions based on variable input — which tool to use, when to hand off to a human. Most projects combine both: the agent decides, the n8n flow executes.
Can our own team update the agent after go-live?
Small adjustments — a prompt update, a new rule — your team can make on its own using the handover documentation. For adding a new capability or extending the architecture, we continue with a monthly maintenance retainer.
Last updated: 7 August 2026
Tell Us About Your Agent Needs