Light follows the work.
Soft color builds around an active worker. Nearby connections flex together, then settle when the work pauses.
THE OBSERVATORY / OPEN TO EVERYONE
AI in motion.
Work you can see.
An evolving network of agents, workflows, pipelines and completed work. See what is happening across your projects—and explore the story behind every new connection.
Created by HillPhantomSupport the ProjectYour support helps keep my software free.
ONE CONNECTED NETWORK. MANY THREADS OF WORK.MAKER NODE / INTERACTIVE DEMONSTRATION
01 / A VISUAL LANGUAGE FOR WORK
More than movement.
Meaning.
The network is a view of activity, not a picture of an AI’s brain. Every visual signal has a simple purpose.
Results become structure.
New connections form as completed work is recorded. A brief label tells you who finished and what they produced.
Every worker has a story.
Open a workspace for its goal, current task, history and outputs. Missions bring the larger picture into focus.
02 / MEET MAKER NODE
A workshop of ideas.
Maker Node is one example: four workers exploring printing, electronics, woodworking and repair. Your setup can use your own agents, workflows or pipelines, across one project or many.
● MAKER NODE / RUNNING EXAMPLEOPEN FULL EXPERIENCE ↗
This public example uses fictional tasks, journals and outputs generated in your browser. It does not run AI, browse sources or train a model. “Live” describes the running demonstration.
03 / YOUR OWN OBSERVATORY
Start with the view.
Connect your work.
Your work, your tools, your structure. Track twelve agents on one project, twelve pipelines, or workers spread across different projects and local AI machines. Twelve is an example—not a role template or a fixed limit.
Download starter ZIP ↓Static browser app · no build step · includes a local preview server, event adapter, sample activity and full instructions.
Give the starter to your AI assistant.
Download the ZIP and upload it to a coding assistant that supports files, or extract it and open the folder in your coding assistant. Tell it what you already run—or what you plan to build—and use the first prompt below. You do not need to adopt a new agent framework. Open app/index.html to explore the example. For your own activity feed, run:
python serve.py # Then open http://127.0.0.1:4182/app/
Give your work a common language.
Define your projects and workers in app/atlas-config.js. A worker can represent an agent, a workflow, a pipeline, or another unit of work you want to follow. Choose your own names, IDs, goals and machine labels; the four Maker Node roles are only the example. Your assistant can add or remove entries as your setup changes. Report stable event IDs, real timestamps and explicit states. Growth follows completed work, not a timer or a token stream.
{
"id": "workshop-event-001",
"worker": "worker-01",
"project": "my-project",
"state": "completed",
"title": "Indexed pegboard holder references",
"topic": "3D printing",
"at": "2026-01-01T12:00:00Z"
}Connect locally first.
Ask your assistant to connect the work you already have through an adapter, then follow CONNECTING.md. The starter reads reported activity; it does not replace or take over your workers. Its external project selector shows all configured projects or one at a time, including workers assigned to different machines. It starts with no invented history or progress.
Keep credentials in the local adapter. Never paste API keys into browser files or a public website.
Share only what you choose.
The demonstration can be hosted as static files without port forwarding. For real public activity, your local adapter would send a sanitized feed outward to a hosted service. That relay is a separate component you must configure; this starter does not expose your computer.
Your domain points to the host. It does not need to point at your AI machine. See the included DEPLOY.md for the deployment checklist.
04 / BUILD WITH YOUR AI ASSISTANT
Upload the starter.
Describe your work.
Give your AI coding assistant the ZIP or extracted folder, explain your agents, workflows, projects and machines, then copy a prompt below. Use it for existing work or work you build next. Ask it to verify the result before calling the feed live.
WHY I BUILT NEURAL ATLAS
HillPhantom ↗I wanted to see my AI work.
I built Neural Atlas for my local AI stack. I wanted a clear view of my agents, workflows and pipelines without rebuilding them around someone else’s framework.
Logs told me what happened. I wanted to see what was happening: what was working, what had finished, and where things were slowing down or going quiet. A living map made that work easier to follow—and helped me understand what it was doing and why.
That is the idea behind Neural Atlas: bring the work you already have, or the work you build next, and give it a visual home. Accurate health and slowdown signals come from the activity your systems report.
COMING SOON / PROJECT ASSIGNMENT
Different projects.
The same local AI.
The project selector is the foundation for following multiple projects, instances and AI machines in one Matrix. Today, the starter filters reported activity by project. It does not deploy projects or redirect compute.
Next, the goal is to assign available local AI resources from the site: let a machine work on Maker overnight, then switch it to Project X in the morning. That means explicit controls and safe handoffs between workloads—not silently stopping running jobs. Workload assignment and orchestration are planned, not included in this starter.
05 / A FEW THINGS TO KNOW
Built to be understood.
Does Neural Atlas run or train an AI?
No. It visualizes reported work. Your AI service runs separately. A journal describes recorded activity; it is not evidence that model weights changed or that a claim is true.
How do I change it to fit my work?
Upload the ZIP to your AI coding assistant or open the extracted folder. Describe the agents, workflows or pipelines you want to track, their goals, project assignments and machines. Ask it to edit app/atlas-config.js, connect your event sources and test empty, running, completed, failed and disconnected states. See CUSTOMIZE.md for a twelve-worker example. No four-role mapping is required.
Can I track multiple projects or AI machines?
Yes: register workers under project IDs and give them machine labels. In external mode, choose All projects or one project from the selector. This is a view of reported activity, not a deployment control. Running or assigning jobs from the site is coming soon.
How do I deploy my version?
Ask your coding assistant to follow DEPLOY.md. The fictional demo can use static hosting and your own domain without port forwarding. A real public feed needs a separate authenticated outbound relay that publishes only approved information. Keep the first real connection local and verify it before sharing.
Can I connect any local AI?
Any workflow that can expose useful task events can potentially be adapted. The starter includes a small event contract and local receiver, not prebuilt integrations for every framework. A token stream alone does not tell you a task completed.
Does the network grow forever?
No. Rendering is bounded: older visual patches are reused. The artwork is a visual window, while your actual system should retain the durable record. The starter keeps a limited session history and does not replace a database.
Is Maker Node really researching?
This public Maker Node is a scripted demonstration. Its sample observations, outputs and daily reflections are fictional. They show how your own completed jobs could appear without sharing private project activity.
What does the starter include?
The working visualizer, Maker Node scenarios, an external-event adapter, example activity, a local-only preview server, setup documentation and copyable prompts. It does not include an AI model, model training pipeline, public relay or automatic daily reflection service.
MAKE SPACE FOR YOUR NEXT IDEA
Your work. In motion.
Take the starter. Keep it local. Make it your own.