Free and open-source AI agents: building an assistant without vendor lock-in
You can deploy a free or open-source AI agent by combining a self-hosted orchestrator (n8n, LangGraph), open-source RAG components, open models (Mistral) and documented skills published under open licences.
The case studies below showcase real deployments built with open components and freely reusable skills.
See also the 1,400+ free Agent Skills and create an AI agent and automate tasks with AI and all use cases.
Open-source architectures, self-hosted n8n and Mistral models
- How Crédit Mutuel Arkéa answers banking questions with Tybot
Crédit Mutuel Arkéa built Tybot on Tock, an open-source conversational toolkit launched by SNCF. The bank's chatbot looks for answers in the customer knowledge base and product sheets.
- How Claire Pro turns PDF orders into carts with n8n and Mistral
Claire Pro, part of Groupe Claire, which has more than 320 employees, receives about 100 PDF orders a day. An n8n agent uses Mistral in a private tenant to extract product codes, quantities and prices. The sales administration team checks the lines before the online store generates the cart.
- How COVERIS logs hours by voice in Excel with n8n and Mistral
At COVERIS, which installs glass roofs and facades, installers report their hours by voice in Teams. n8n, Mistral/Voxtral and Excel transcribe and structure the entries and produce the monthly summary for accounting. It takes 30 seconds per entry and about 10 minutes of checking each week.
- How Fayat Immobilier triages warranty defects in two minutes
Fayat Immobilier and FlowMinds deployed an n8n and Mistral agent for defects reported under France's guarantee of perfect completion (GPA). It extracts details, suggests a severity rating under the general contract conditions (CCAG), and drafts a work sheet and acknowledgment. Processing went from 25 minutes to about 2 minutes, a 92% drop and over 100 hours saved per manager per year, with human approval.
- Scan job boards each morning with n8n and Claude
At BCP Partners, a Paris executive recruitment firm, an n8n workflow scans job boards and career sites each morning. Claude Haiku extracts the company and role, seniority rules sort the results, and an email digest arrives at 8:45 a.m., saving around 2.5 to 4 hours a week.
In brief, with the sources.
Are there genuinely free and open-source AI agents?
Yes: orchestrators like n8n (Community edition) and frameworks like LangGraph are open source, and the 1,400+ Agent Skills in the /skills/ directory can be inspected on GitHub under their respective licences.
Can you deploy an open-source AI agent on internal company data?
Yes, as demonstrated by Arkéa with an open-source RAG architecture and several French SMEs running n8n with Mistral.
Related themes
Other guides
- Create an AI agent
- AI agents for small businesses
- AI agents at work
- AI for admin procedures
- MCP server (Model Context Protocol)
- ChatGPT apps
- Automate tasks with AI
- Agentic AI
- AI agents for freelancers in France
- Claude Code: uses and alternatives
- AI agent platforms & French directories
- AI commercial film production
- Which AI subscription to choose?
- Our editorial method