Article reviewed and updated by our Atlassian and AI experts on 8 December 2025.
A few essential points to remember:
-
BleuLemon is not a software vendor: the firm supports innovation inside IT departments.
-
In a culture of permanent innovation, our experts prototype mock-ups to align needs with AI features, then build, test and deploy them in real conditions.
-
The conversational AI agent developed by BleuLemon simplifies the user experience of the service portals offered by Jira Service Management.
-
The solution rests on a hosted multi-agent AI architecture, which guarantees complete control over the data.
-
The agentic solution is independent of the vendor, so that IT teams keep their hand on maintenance and future changes.
-
Alongside Atlassian's Rovo, the BleuLemon approach offers an independent AI, compatible with Cloud and Data Center, and controlled from end to end.
To cut the time spent on Atlassian's Jira Service Management portals, the BleuLemon teams developed a proof of concept (PoC) for an intelligent conversational agent. Built on a multi-agent architecture, this project simplifies the user experience and guarantees independent data management.
I. Simplifying the JSM experience
Jira Service Management (JSM) portals from Atlassian are powerful, but they are often dreaded for their complexity. Between the forms, the many sub-menus and the redundant options, navigating a JSM portal can quickly turn into an obstacle course.
“Personally, when I see that kind of interface, it makes me want to close it straight away. I was looking to lighten the actions and make them safer”
Emmanuel Mancuso, consultant and data scientist at BleuLemon.
Behind the apparent simplicity of the AI chatbot, the whole complexity of the JSM architecture is hidden away to make room for a conversational AI agent developed by BleuLemon, able to understand requests in natural language and answer them in a few seconds.
II. An independent, fully controlled AI architecture
Under the bonnet, the PoC rests on a multi-agent architecture connected to an MCP (Model Context Protocol) server developed by Souha Kassab and Emmanuel Mancuso , AI experts at BleuLemon .
Each agent has a precise role: some interpret the request, others interact with the Jira APIs through the MCP server, others handle the rewording or the validation of the information.
The real singularity of this PoC lies in the technology choice: GPT-OSS models hosted on OVHcloud, which guarantees that the data never leaves the infrastructure controlled by BleuLemon.
“The processing pipeline is fully controlled”, explains Emmanuel Mancuso, an approach radically different from proprietary AI solutions, where requests can be used for training purposes by third parties.
Case study: anonymising health data with an AI agent
In a healthcare context where data has to be anonymised, we carried out a bulk import into Jira Data Center and ran a Cloud migration audit, to secure the data and check GDPR compliance.
We developed an AI agent that automatically monitors the attachments on tickets. Every time a PDF or image is added, the agent reads the content, spots sensitive information (for example a social security number, an IBAN, names) then masks or replaces it before the document circulates or is reused. In other words, the clean-up happens as it goes, with no manual action, to avoid any accidental leak.
This first agent let us prove that the value of AI agents which automate such checks comes both from their orchestration and from clearly defined governance.
The IT department stays responsible for what it wants to anonymise and when.
BleuLemon hands over the AI PoC solution and the operating procedures so that your team can run it on its own.
III. BleuLemon support: from advice to deployment
This technological independence is part of a wider vision: BleuLemon is not a software vendor.
The team delivers an operational, working version of the AI agent, then supports its clients so that they can take over the maintenance and the development of the solution.
This digital independence is a key argument with large accounts, for whom confidentiality and control of their environments are essential prerequisites.
“We want companies to stay autonomous and independent in the management of their AI governance”, says Stéphane Génin, President of BleuLemon:
“They can rely on our expertise for scoping, integration and security, but without depending on a subscription or a third-party vendor.”
This approach reinforces the philosophy of the project: no proprietary lock-in and independence from end to end.
-
Sovereign, controlled AI
Your data stays with you. Independent models. Compatible with Atlassian Cloud and Data Center. -
Fast deployment In production in 2 months. Training for your teams included.
-
Complete autonomy No recurring subscription. You manage the maintenance. No proprietary lock-in.
-
Measurable ROI Shorter processing times. Running cost: a few euros a month.
IV. From technology to strategy: an advantage over Atlassian's Rovo
The agentic PoC built in October 2025 by BleuLemon sits inside an ecosystem Atlassian has already explored through Rovo, its own AI assistant.
BleuLemon offers an agentic AI solution compatible with Cloud and Data Center, but above all independent and controlled from end to end.
This strategic direction reflects a clear intention: not to depend on a single technology player, and to give clients the option of applying their own security policies.
It is a philosophy that resonates strongly at a time when governance and the transparency of algorithms are becoming priorities.
V. Which technical challenges had to be overcome to develop this AI agent?
Developing this PoC was not a smooth ride.
The BleuLemon team had to overcome several major technical obstacles, starting with model hallucinations, when the AI generates wrong answers.
“The agent hallucinated completely”, recalls Emmanuel Mancuso, looking back at the first tests. The problem was gradually brought under control through precise prompt engineering and by specialising the agents in clearly defined roles.
Another challenge was handling authentication on Jira Cloud and Data Center through OAuth 2.0, essential to secure the exchanges between the systems.
This independent technology approach called for specific work to keep compatibility with both Atlassian's Cloud environments and the clients' internal infrastructure.
Despite these constraints, the results are there: the conversational agent answers requests in a few seconds, offers direct links to the right forms and guides the user through creating requests .
All of it for a minimal cost: a few euros a month for millions of tokens consumed.
VI. Human in the Loop: help with creating requests, under user control
Beyond the technical feat, BleuLemon's ambition does not stop at automating actions in Jira Service Management. The PoC was designed to help the user create their request, while keeping one key principle: the human stays the decision-maker.
In practice, BleuLemon put in place a Human in the Loop logic in the form of help with creating a request: the conversational agent helps qualify the need, offers a structured rewording, pre-fills the necessary fields (category, description, priority, context elements), then submits a clear summary to the user.
Before the ticket is actually created, the user can:
-
validate the information offered,
-
correct or complete certain fields,
-
adjust the level of detail,
-
or cancel the submission.
This approach guarantees safety in the use of AI, where automation takes on the repetitive, time-consuming tasks, while the IT support teams keep oversight and the final decision.
As Stéphane Génin, president of BleuLemon, sums it up:
“Generative AI takes on the complexity, the human keeps control.”
Conclusion
The experiment run by BleuLemon around Jira Service Management illustrates the new generation of AI integrations: transparent, independent and centred on the user.
By proving that technological performance and respect for data can be reconciled, this PoC opens the way to a lasting transformation: an AI that assists without dominating, and that finally makes technology more human.
