Project Fusion

Your AI Doesn’t Know Notes. Yet.

Project Fusion* connects the AI you choose to your HCL Domino data. Under each user’s own rights, on your own server, with a record of what happened.

* Codename. The product’s name comes later.

We are building Project Fusion for enterprise demands, and sizing it for any estate: from a single server with 50 users to large enterprises, and everything in between.

  • Single sign-on
  • Audit record
  • Control on top of your ACLs

01.

Your rights, and your rules on top

Each user signs in through your own identity provider, and Domino’s ACLs, roles and Readers fields apply, as in Notes. On top, you can narrow what each user or group may reach through the AI.

02.

Made for audits

You see who used the AI, on which databases, and how much data went back to it. Every answer can be traced and checked: the trail your auditors and your DPO ask for.

03.

Learns your systems

The more your teams use it, the better it knows your applications. What worked once is kept and shared, on your server, whichever AI you use.

04.

Your AI, your server

Any AI whose client connects to MCP servers with sign-in.* Proven today with Claude, Microsoft 365 Copilot and GitHub Copilot, on the web, the desktop and the phone. Or a model you run yourself. You run Project Fusion on your own Domino server. panagenda hosts nothing.

* The AI’s client must support enough of MCP. We publish the requirements. Microsoft 365 Copilot needs an agent set up in your tenant. A model you run yourself works through any client that does.

Be the first to know

What’s new, and when version 1.0 arrives.

Measured, not claimed

One question: “Which databases can I read, and with what access?” Across 66 databases, counted in tool calls.

One tool call per step, the conventional MCP way

67 calls

Project Fusion, the first time: the AI writes one small script

2–3 calls

Project Fusion, asked again: the saved script, no AI tokens to run it

1 call

The bigger the question, the bigger the gain. On a one-line read, both ways cost about the same.

Our own bench, September 2026: Claude Sonnet and Opus on our test system, 66 databases. The conventional way is one tool call per operation on our own core, not a test of other vendors’ products. The times there: 90 to 120 seconds, 40 to 70 seconds, and 10 to 30 seconds asked again; they depend on the question, the model and the server.

How it works

You ask.

In plain language, in the AI you already use.

You sign in.

Your AI, in its vendor’s cloud or on your own hardware, connects to Project Fusion over HTTPS, as you, through your identity provider.

The work happens next to the data.

The AI writes a small script. Project Fusion runs it on your Domino server, with your rights and within your rules.

Only the result goes back.

Not whole views. The script is kept, so it can be checked and run again.

Say yes to AI, and keep control

Your rules apply

  • Every request runs as the person who asked. No shared token, no service account.
  • ACLs, roles and Readers fields, exactly as in Notes.
  • A second layer on top: narrow, per user, group and database, what may be reached through the AI. Someone who opens a database in Notes need not reach it through the AI. This layer only takes away: it never grants more than the ACL does.
  • Reads first. The first version reads only. Writing comes later, and only where you switch it on, under your rules.

You stay in control

  • An audit record: who used the AI, on which databases, how much data went back, and what was exported.
  • You choose the AI. The access path stays under your control. What goes back to your AI vendor is reduced on your server first. With a model you run yourself, it stays in your house.
  • Safeguards you set. Every run has limits, for the server and for each user, and the audit record shows how much went back.
  • You decide when to update. OnlineUpdate tells you a new release is out, and your admin installs it when it suits you.

Built with the rules you work under in mind: GDPR, the AI Act, NIS2 and the Cyber Resilience Act.

Licensing

A subscription per user. No token fees, no per-call charges. Prices come at launch.

Platforms

HCL Domino 12.0.2 or later, on Windows and Linux.

Start with what you have

Your Domino servers and your AI, such as Microsoft 365 Copilot. No new hardware, no other new software.

Reads, then writes

The first version reads only. Writes come later, if you allow them.

Availability

Version 1.0 planned for early 2027.

Frequently asked questions

Which AI works with Project Fusion?
Why MCP?
Do we need the Notes client?
Does it respect our ACLs and Readers fields?
Can we limit what people do through the AI?
Can we audit what the AI did?
Does our data leave the house?
Can it change our data?
Is it a Notes agent, or LotusScript?
How does the AI reach our server?
Which Domino version do we need?
What does it need on our server?
How fast is it?
Is it only for large organizations?
We have many Domino servers. Do we need one per server?
How is it different from iDNA Companion?
And HCL Domino IQ?
What does it cost?
When can we have it?
Can we join the preview?
What will the product be called?

From the team behind panagenda MarvelClient,
with decades of experience on HCL Notes and Domino.

Intelligent HCL Notes Management Doesn’t Stop Here