Hermes Agent training: an AI agent that runs on your own machines
You want an agent that works without depending on a vendor, a subscription or a cloud. Hermes Agent is open source, under the MIT licence, and runs on your teams’ own machines. In one Qualiopi-certified day, your technical people install it, give it real work, write reusable procedures and set the guardrails.
In-house training, on-site or remote, for one to ten participants. Price per group: €1,000 to €2,000 excl. VAT for half a day, €1,800 to €4,000 excl. VAT for one day, €3,500 to €7,000 excl. VAT for two days. OPCO funding possible.
Colombani.ai offers a one-day programme on Hermes Agent, the open-source AI agent from Nous Research under the MIT licence, Qualiopi-certified and eligible for OPCO funding. Participants install the agent and a local model on their own machine, give it a real task from their team, write a reusable skill, schedule an automation and draft their team’s usage note. One thread runs through the day: at every stage, which data leaves the machine, to whom, and in what form. The executive briefing, beforehand, is where the choice between a local agent and a hosted assistant is made.
One programme per audience, delivered on Hermes
Each programme is part of the Qualiopi catalogue and uses your own cases: your tasks, your files, your machines. Participants install it themselves and leave with an agent that runs.
Running an open-source AI agent on your own machines
Installing the agent and the local model to suit the hardware, a first work task run end to end, reusable skills, a recurring automation, then the guardrails: approval, access separation, logging, a pinned version.
- Installation and local models
- Skills and profiles
- Guardrails and logging
AI training for executives and boards
Before deploying: what a local open-source agent brings and what it costs, compared with a hosted assistant. The five main market tools compared (Claude, ChatGPT, Microsoft Copilot, Gemini and Mistral Vibe) on your cases, the regulatory framework, the data questions, and a prioritised action plan to take away.
- Local or hosted
- Use cases by function
- AI Act and data
Open source makes the agent readable; training makes it usable
An open-source agent has two qualities hosted assistants lack: it runs on your machine, and its procedures are files you can read, edit and version. Hermes Agent, published by Nous Research under the MIT licence, has both. That alone does not make it fit for professional use: the day is what takes you from an installation that works to an agent a team can be handed, with an approval rule, separated access and a log.
Advice remains independent of vendors. Colombani.ai also deploys Claude, ChatGPT and Vibe on engagements, and says so when a hosted model is the right call for a given case. What your teams learn here, scoping a prompt, writing a procedure, checking a result, setting a guardrail, applies to any agent, whoever publishes it.
Local AI training →From installation to day-to-day operation
-
Installation and local model
Installing the agent in desktop mode, reading your hardware, choosing a sensible model size, comparing two models on the same work prompt.
-
Skills and profiles
Turning a procedure into a reusable skill, and knowing when a need warrants a separate agent profile rather than one more skill.
-
Automation
Background tasks, driving the browser on an internal application, choosing the surface: desktop, internal chat, command line.
-
What leaves the machine
Local model, tool gateway, web search, session summarisation: the list is kept through the day and checked against network traces, not assumed.
Qualiopi-certified provider, OPCO funding possible
Colombani.ai is a Qualiopi-certified training provider (certificate QUA009472, training provider registration number 11757549975). These programmes are therefore eligible for OPCO funding, or FIF-PL funding for self-employed professionals, subject to the funding body's decision. The tool is chosen during the needs analysis and stated in the training agreement.
How to fund AI training through an OPCO →
Ulysse Trin
Founder of Colombani.ai, holder of the Claude Certified Architect Professional certification issued by Anthropic and of certifications issued by OpenAI
Engineer (Epitech, MSc AI), holder of the Claude Certified Architect Professional certification issued by Anthropic and the ChatGPT Deployment Practitioner certification issued by OpenAI, both since 19 September 2026. Consultant, trainer and developer: every week on client engagements he uses Claude, ChatGPT, Codex, Vibe, Copilot and Gemini, and his programmes teach only what he has built there.
Claude Certified ArchitectProfessional · Official Anthropic certification. Verify the credentialand nine practitioner certifications issued by OpenAI → Is a subscription needed to attend?
No. The agent and the models used are open source, the agent under the MIT licence. A subscription to a hosted model is optional and only serves the comparison made at the start of the day, to establish what leaves the machine in each case.
What hardware is required?
An Apple Silicon Mac or a recent PC, at least 16 GB of RAM, 30 GB of free disk space and administrator access. If the machines are managed by IT, installation permission is settled before the session. Hardware is checked during the needs analysis, not on the day.
The tool moves fast. Will the materials be current?
The version used is pinned for the session and recorded in the report; the materials match that version. That is also what the day recommends for production use, and running an upgrade is part of the programme.
Is the programme adapted to the tool or generic?
Adapted. The learning objectives and assessment are those of the programme in the Qualiopi catalogue, which makes it possible to request funding, subject to the funder's decision; the materials, the installation guide sent three days before and the exercises are those of the chosen tool, with its own features. One sequence compares the five main market tools (Claude, ChatGPT, Microsoft Copilot, Gemini and Mistral Vibe) from a confidentiality standpoint, whichever one you use.
Can it be funded through the CPF?
No. Colombani.ai programmes are not listed for CPF funding. As Colombani.ai is Qualiopi-certified, funding can be requested from your OPCO or the FIF-PL; the decision rests with the funder, according to its criteria and budgets.
Open enrolment or in-house training, on-site or remote?
In-house training, using your real cases and data, for one to ten participants. On-site in France or remote, as you prefer. An open-enrolment format is occasionally offered for the executive briefing.
Need training on an open-source agent?
Describe your teams, your use cases and the tool you use. Receive a response within 48 hours with a programme and a quote.