I build AI agents in Python that read a request, look up real data in your systems, and take the right action: update a record, create a ticket, draft a reply or start a workflow. Your rules decide what the agent may do, and anything risky waits for a person to approve it.

This is for you if • Your team handles the same requests again and again: look something up, then update a system • You want AI in your workflow but worry about it making costly mistakes • You use n8n or your own APIs and want an AI step that's reliable • You tried a chatbot and now need one that can safely take actions

What you receive • An AI agent that understands the request and picks the right tool for each step • Tools connected to your systems: your API, database, CRM, e-mail, Google Sheets or Slack • Answers based on real data looked up first, never guessed • Human approval for risky actions such as refunds, deletions or sending messages • Hand-off to a person when the agent isn't sure • A log of every decision and action, so you can always see what happened and why • Full Python source code (FastAPI), automated tests, Docker and a setup guide

Options you can add • AI models: OpenAI or Claude, or an offline model for sensitive data • Workflows: self-hosted n8n workflows that trigger the agent or act on its results • Knowledge: answers grounded in your own documents (RAG), with sources • Channels: a web chat, Slack, e-mail or your own app • Safety: prompt-injection screening, personal data masked before the model, daily cost budgets • Operations: an approval queue for your team, plus usage and cost reports

You describe the task, the systems involved and what the agent may and may not do.

I send a written plan: tools, approval rules, data access, and what's in and out. Nothing starts until you approve it.

I build the agent and test it on real examples, including tricky and malicious inputs.

We run it together in a safe mode first, with every action requiring approval.

You receive the code, the guide and a walkthrough, and I make your changes.

Proof you can check My open-source AI support agent has 13 tools, each allowed only for the right kind of request, with budgets, timeouts, an audit trail and human approval for refunds and cancellations. It resists prompt injection and has 484 automated tests at 90% coverage. https://github.com/mojtaba-py-code/secure-ai-customer-support-platform

NexusFlow, my automation platform, runs five n8n workflows with Python execution, a service token per workflow and an emergency stop switch. https://github.com/mojtaba-py-code/fastapi-postgres-ai-automation-platform

Not included Agents that move money or delete data without a person's approval, Make or Zapier builds, and paid AI or API usage (billed to you directly by the providers).

Ready to start? Send me a message describing the task you'd like an agent to handle, and I'll reply with a plan before you order.