[ In short ]
An AI agent is software that uses a large language model to complete a real task, such as answering a customer, qualifying a lead or filing a document, using your tools and data. Lunarion Labs builds custom AI agents with guardrails, evals and human approval, for US companies and software products.
What are we talking about?
The model is the easy part. What makes an agent useful is integration: access to the right data (often through retrieval, called RAG), permission to use the right tools, clear rules, and human review on sensitive actions. Useful projects start from one measurable task, not from a demo.
Most companies have tried AI. Few have connected it to the systems where the work actually happens. The value is in the integration: the right data, the right tools, clear rules and human approval on anything sensitive. We build agents that complete real tasks and we measure them in hours saved.
When should you call us?
- Your team answers the same questions all day.
- Data from documents is retyped by hand.
- You tried AI but never connected it to your tools.
- Your data cannot leave for a third-party model.
How does an engagement work?
- Pick the task. One repetitive, measurable task with a clear payoff.
- Prototype. A first agent tested on your real data.
- Guard. Evals, rules and human approval.
- Deploy. Connected to your tools, with quality and cost tracking.
The starting point is always the same: the Lunarion audit, at a fixed price, which gives you a written, costed diagnosis. See our four-step approach.
AI & agents: what is included
- Task agents : Agents that act: qualify a lead, draft a quote, update a CRM record, route a ticket.
- AI customer service agents : Answers on your site, in email or chat, with a clean hand-off to your team when needed.
- Knowledge search (RAG) : Ask questions across contracts, policies and docs, and get sourced answers from your own data.
- AI features in your product : Summaries, classification, search and generation built into your SaaS or app.
- Document processing : Invoices, forms and emails read, extracted and filed automatically.
- Connected to your stack : HubSpot, Salesforce, Slack, Gmail, Zendesk, your database: the agent works where your team works.
- Evals and guardrails : Answer-quality tests, rules and human review before an agent goes live, and after.
- Private and compliant AI : Model and hosting choices that keep sensitive data inside your perimeter.
What drives the cost?
Every project is costed after the audit, which has a fixed price. The main factors are:
- The target task and its volume
- The data sources to connect
- The model and its per-use cost
- Privacy requirements (hosted model or API provider)
Mistakes to avoid
- Launching a generic chatbot with no company data
- Letting an agent act on sensitive data without approval
- Shipping without evals to measure answer quality
- Sending confidential data to a provider without the right settings
Where do we work?
Lunarion Labs is a software lab based in New York that works fully remotely, in English and French. We work with companies in the US, France and Spain, including in these cities:
United States
New York, Los Angeles, Chicago, Houston, Dallas, Austin, Miami, San Francisco, Boston, Seattle.
France
Paris, Lyon, Marseille, Toulouse, Bordeaux, Lille, Nantes, Strasbourg, Nice, Montpellier.
Spain
Madrid, Barcelona, Valencia, Seville, Malaga, Bilbao.
Go further
- Automation : to chain AI with your business workflows.
- SaaS & business software : to build AI features into your product.
- SEO & AI visibility : to get cited by AI assistants.
- Software project rescue: rewrite vs refactor, and how to decide
