Development

AI agents, plugins and integrations, delivered with their code.

AI agents custom-built for your processes, plugins for your teams' tools, AI integrated into your products, and products developed and operated by Colombani.ai. Each solution is delivered with its code, tests and documentation.

Book a call

See our products

Our services

Agents, plugins, security and integrations

Agents

Custom AI agent for a business process

One specific business process handed to an agent: it reads the documents, prepares the work and flags what is missing, while decisions stay with your teams. Tests, performance measurements, logging and oversight included.

API Deployment Practitioner certification, issued by OpenAI, for taking to production on the API.

Team trained and first agent Plugins

Business plugins for Claude and Codex

Your teams' procedures, templates and tools packaged as a plugin: skills that encode your know-how, commands, hooks that set the guardrails, and MCP servers that connect AI to your tools (CRM, document management, databases, internal APIs). Delivered tested, with source code.

Explore the offer AI security

AI pentest and security audit for AI agents

An audit of your AI agents and the applications built around them: prompt injection, data leakage, tool misuse. Carried out with your written authorisation, within a written scope, with every finding validated by a person.

Explore the offer Integration

AI built into your products

For software vendors and product teams: integrating a language model into your application without depending on a single provider, with control over the cost per request, data isolation between customers and a suitable data processing agreement.

Software and SaaS vendors

Tech stack

Claude API / Agent SDK Core AI engine. Reasoning, generation, analysis.
Skills and plugins Your teams' know-how encoded and shared in Claude, Claude Code or Codex.
MCP (Model Context Protocol) Connection to business tools: CRM, databases, internal APIs.
Local models (Ollama, Mistral) When data must stay on your servers.
Python / TypeScript Backend and automation. Every project is delivered with its source code.

An agent that prepares your case files

It reads incoming documents, extracts what matters, prepares the first draft and flags what is missing. The professional reviews the output and makes decisions instead of re-entering data. Timeline set at scoping, source code included.
An agent that handles incoming requests

It reads emails and forms, assesses the request, routes it to the right person and drafts the reply. Your team only handles requests that require a decision.
An assessment of what your AI can access

Which data deployed assistants can read, which connectors are open, and which access rights remain active after a departure. A prioritised report and remediation plan.
A penetration test of your exposed systems

Attack surface mapping and vulnerability detection accelerated by AI agents and validated by a person; controlled exploitation entrusted to a specialist offensive security provider, coordinated by Colombani.ai. A report and supporting evidence delivered.

In-house products

Our products

Tools designed to meet needs identified in practice, developed and maintained by Colombani.ai. They illustrate what a well-scoped AI agent can accomplish.

Personne: original and redacted copy side by side, fictional data

Personne: dropping a case file to prepare

Prototype 0.5 · Mac with Apple Silicon and Windows

Personne

Your case file redacted, on your own computer.

A local application that prepares a case file before it is shared with an AI agent. It identifies names, organisations, addresses, email addresses, phone numbers and IBANs, and replaces each with a consistent label such as [PERSON_1]. You review and approve each document. It then exports text or PDF copies that Claude, Codex or Cowork can read. The originals remain intact, no text leaves your computer, and the AI's response is restored with the original values on your own machine.

Local engine with no network calls and built-in detection models
Labels consistent across documents, explicit approval before export
Redacted PDFs with content and metadata permanently removed
Original values restored in the AI output, locally

For law firms, notarial offices, accounting practices, HR and any team handling files containing personal information.

Prototype currently being tested. No detection process identifies everything, so review remains essential. Pilot users wanted.

Request a demo

Genomix

Genomic analysis in natural language

A tool for biologists and researchers. They ask a question about a genetic variant file. The agent reads the file, queries thirteen reference databases and explains the results. It runs locally and can be installed with a single command.

Python
Claude Code
MCP
Ollama

View on GitHub ↗

France Dashboard

Citizen portal for public data

Elections, economy, security, energy, health, education: fifteen open datasets, 450,000 rows. Ask a question in French and the agent generates the query and chart. Live at francedashboard.fr.

React
FastAPI
PostgreSQL
LLM

Visit the site ↗

Method

A scope written before the first line of code

01

Scoping

What problem, what constraints, what tools in place. Deliverable: scoping document.
02

Proposal

Technology choices, scope, timeline, budget. Deliverable: a detailed proposal.
03

Build

2-week sprints, demo at each iteration. Deliverable: solution tested in your environment.
04

Follow-up

Fixes, enhancements, handover. Deliverable: documentation and 30 days of support.

Sector-specific approach

Agents and audits tailored to your profession

Professional secrecy, HDS hosting, professional ethics and stricter GDPR requirements are incorporated into the agents and audit scopes. Explore the approach for each regulated sector.

Explore the sectors

FAQ

Frequently asked questions

How much does a custom AI solution cost?

Each project is priced individually after a free scoping discussion. The scope depends on complexity, integrations and the required level of refinement.

Who owns the code?

You do. Source code, documentation and access credentials are delivered at the end of the project. The delivered work is yours, with no recurring licence fees.

Can we use local AI for sensitive data?

Yes. Colombani.ai deploys local models (Mistral, Qwen via Ollama) directly on your servers or workstations. No data leaves your infrastructure. See the sovereign architecture offering.

How long does it take to build an agent?

The timeline is set at scoping, according to the agreed scope: the number of tasks given to the agent, the tools to connect and the level of approval required. The schedule is written into the proposal before any work starts.