Ideas should not belong only to people with a team behind them. Coding Sensations is Michael DeMott’s exploration of AI that gives more people the tools, support and freedom to build.
YOUR DIRECTION / A TEAM AROUND IT
One idea. Many minds.
Imagine assistants with different strengths working together around your goals. Mira is our AI assistant character—a friendly face for a bigger idea: useful teamwork shaped by the person it serves.
CONNECTED TOOLS / ROOM TO GROW
Work together. Remember more.
We’re connecting agents, models, memory and creative tools into one adaptable workspace. A task can move between specialists, useful lessons can carry forward, and you can see what is happening.
YOUR WORKSPACE / YOUR CHOICE
Choose the tools. Shape the experience.
A desktop workspace. A phone. A companion beside you. We’re building toward a system where the way you interact—and the intelligence behind it—can change with you.
THE BIGGER GOAL
More agency. More possibility.
People with ideas deserve more ways to act on them. We want to make powerful tools approachable, adaptable and personal, so more people can create things they could not build alone.
0105
KEEP EXPLORING ↓
ONE CONNECTED WORKSPACEEXPLORE THE SYSTEM
One idea. A whole system.
We’re building an adaptable workspace where agents collaborate, useful knowledge carries forward and people choose how to connect. Select a module to explore its part of the system.
↶ Knowledge and reviewed improvements feed the next task. Companions offer another way into the workspace.
This is the integration vision. The modules combine implemented components with work still in development; each expanded flow includes its current progress and upstream credit.
Choose any module above to see its detailed flow.
03 / HOMELAB / CONNECTED WORKSPACE
One place to return to. Many ways to connect.
HomeLab brings the pieces closer to the person using them: a local-first workspace spanning native Apple and browser interfaces, computers and assistant connections. The long-term goal is a consistent place to continue your work, whichever interface or assistant you choose.
The workspace / connected interfaces
You & your devicesNative app or browser.
HomeLabHistory, drafts and connections.
Assistant environmentHermes and other supported routes.
Useful outcomesWork, knowledge and companion experiences.
↔ Context and progress return to the workspace. Devices connect deliberately; existing home automation remains independent.
Continuity across interfaces
Keep useful history and recoverable drafts close to the work. Moving between interfaces should feel like continuing a project, with the source of each assistant’s contribution still clear.
A common home for integrations
Adapters connect different assistant interfaces to a shared experience. That makes HomeLab the meeting point for the agent environment, inter-agent communication and future companion connections.
Infrastructure that serves the idea
The local foundation lets us explore connected tools without making the whole experience depend on one provider. It also leaves room for the practical realities of reconnects, device changes and interrupted work.
Native client, browser fallback and gateway components are implemented. Live cross-device and companion connections remain in development; HomeLab is presented here as a private engineering project.
01 / HERMES AGENT / INDEPENDENT FORK
A home for agents that work together.
We’re extending Hermes into a configurable environment for autonomous work: profiles with distinct roles, bot groups that bring them together, and model choices that stay in your hands. The goal is a personal team that can keep moving while you retain direction and insight.
The agent environment / how work takes shape
Your goalSet the direction and project context.
Profiles & groupsBring the right roles together.
Models & toolsChoose how and where work runs.
Results & insightFollow progress and guide what happens next.
Local and cloud routes serve different needs. Egress controls inspect supported model requests and destinations; model choice alone is not a privacy guarantee.
Purpose-built teams
A planner, builder and reviewer can have different instructions, tools and working context. Profiles and bot groups provide the structure for those roles; our integration work connects that structure to real project work.
Freedom across providers
The vision is to choose a model for the job without rebuilding the whole experience around one vendor. Local execution and deliberate cloud routing let us make explicit choices about capability, context and data leaving the workspace.
Autonomy you can steer
Boards, task history, pause controls and meaningful diagnostics put the work in view. Checks and balances belong inside a capable agent environment, so you can intervene, change direction and understand why something stopped.
Built on Hermes Agent by Nous Research and upstream contributors, including its existing profiles, memory and skills foundation. Michael directs the independent fork’s integration and extension work with AI implementation and review collaborators.
The fork includes routing controls, task coordination changes and regression-tested improvements. Connecting the installed components into the complete cross-machine experience is ongoing.
02 / A2A / INTER-AGENT COMMUNICATION
Different agents. A shared conversation.
A2A is about helping agents work harmoniously. A specialist should be able to receive useful context, contribute its part and return a result that another assistant can build on—across the tools and environments where those assistants already work.
A collaborative task / example integration goal
Plan togetherTurn the goal into useful pieces.
Share contextSend the specialist what it needs.
Do the workContribute through its own tools.
Bring it backCombine results and continue.
↔ Communication carries context and progress between agents. Reconnecting to the conversation should not repeat the work.
A bridge between assistants
Different assistants speak through different interfaces. A2A explores the shared messages and handoffs that let them participate in one effort without requiring them all to be the same program.
Specialists with a common goal
A planning assistant could ask another to investigate an idea while a builder works on implementation. The value is in useful exchange: enough context to contribute and a result the rest of the team can use.
Part of a larger workspace
Hermes supplies an agent environment; A2A connects it to other participants. HomeLab is where a person can follow and guide the work, while selected outcomes can feed the wider memory and review loop.
Communication and handoff components have offline tests. Live connections between all intended assistants are still being brought together.
ACROSS THE SYSTEM / MEMORY & IMPROVEMENT
Let the next project start further ahead.
A useful assistant should build on what you have learned together. We’re bringing Obsidian notes and managed agent memory into the broader workspace so decisions, preferences and useful lessons can remain available beyond one chat.
The continuity and improvement loop
ExperienceWork on a real task.
Useful knowledgeKeep selected notes and lessons.
ReviewDecide what should carry forward.
Better foundationsRefine instructions, skills and tools.
↶ Bring the reviewed knowledge and improvements into future work.
Knowledge you can work with
Obsidian offers a human-readable knowledge space. Managed memory serves the agent’s working context. The aim is to connect these deliberately, so a note can inform future work without every conversation becoming permanent memory.
Character and direction
An agent’s “soul” is the identity and instruction layer that shapes how it behaves. We’re exploring reviewable improvements to that layer, so personality and working habits can evolve with the person and the task.
A better working environment
The harness is everything around the model: tools, memory, routing and execution. Improving that environment can matter as much as switching models. Skills and review workflows give those improvements a repeatable shape.
This builds on Hermes’s memory and skills capabilities and the community Obsidian ecosystem. The integrated learning loop is a development direction; the SkillOpt diagnostic workflow currently uses a mock backend.
SKILLS, PLUGINS & TOOLS
Better tools. Bigger possibilities.
Skills capture how to do a job; plugins connect the capabilities and interfaces that make it possible. Together, they let the workspace grow into software tools, knowledge spaces and creative production.
A reusable workflow / from idea to improvement
Choose a jobSoftware review, creative production or learning.
Connect the toolsPlugins bring capabilities into the workspace.
Use a skillFollow a reusable way of working.
Review & improveCheck the result and carry useful lessons forward.
01 / SOFTWARE REVIEW
Keep feedback useful
Connect review findings to the code they describe, so an assistant can turn feedback into a concrete change and a reviewer can follow what happened.
Progress: a revision-aware GitHub review workflow is implemented. The next platform step is GitLab support.
02 / ARTWORK INTO OBJECTS
Card to 3D
Carry an approved character design from multiview artwork into a scaled Blender starting point, opening a path from a visual idea to a dimensional companion.
Built around Tencent’s Hunyuan3D-2mv and Blender, with artistic cleanup and repeatable reproduction shaping the next iteration.
03 / REVIEWABLE IMPROVEMENT
SkillOpt-Sleep
Explore how experience can become proposals for better skills and working instructions, using Microsoft’s SkillOpt and Hermes as part of the learning-loop direction.
Progress: a diagnostic prototype gives this experiment a concrete starting point. Its backend is a mock; the next step is a real, reviewable improvement cycle.
04 / EXPRESSIVE CHARACTERS
Animation Quality Gate
Give motion a deliberate review process: check framing, silhouette, flicker and exported movement so a character’s personality survives the production pipeline.
Progress: a procedural review workflow gives creative judgment a repeatable structure. The direction is more consistent character and companion production.
05 / PLUGINS & CONNECTIONS
Extend the workspace around you
Bring new tools and interfaces into the agent environment rather than rebuilding the experience for every integration. The direction is a workspace where knowledge, project tools and companion connections can be added as the person’s needs grow.
Connection-aware plugin settings are one step already built into the fork: settings follow the active gateway connection. Community extensions such as Obsidian memory provide foundations to build on, alongside our own integration work.
06 / SELF-HOSTED RUNNERS
A place to build and check
Self-hosted runners provide build and test capacity on our own computers. They support the software behind the agent environment and creative tools, while remaining a separate system from the agents doing project work.
The integration goal is repeatable checks across the intended repositories and platforms, with practical resource limits and recovery. This is infrastructure engineering, not an offer of access to Michael’s machines.
04 / STACKCHAN / CHOOSE YOUR COMPANION
Your companion. Your choice of intelligence.
We’re building toward a choose-your-own companion experience on StackChan: a character with expression and presence, connected to the assistant you want to spend time with. Muse, Hermes, ChatGPT and Claude are part of that integration vision, rather than one permanently fixed personality or backend.
The companion experience / intended direction
Choose a characterShape its look and personality.
Choose an assistantMuse, Hermes, ChatGPT or Claude.
Make a connectionUse a supported integration.
Meet your companionBring conversation and expression to the desk.
Provider and character choices are integration goals; availability depends on the supported route and device work.
A personal presence
A desk companion can make interacting with AI feel more immediate and playful. The goal is to let people choose the character and experience that feels right for them.
Part of your wider world
The companion is another way into the connected workspace. We’re exploring how selected context and capabilities from the broader system can support a coherent experience across screen and desk.
From artwork to interaction
Character navigation, face rendering and lifecycle work form the practical foundation. Creative workflows add another dimension: developing the visual identity and motion that make the companion feel like itself.
An independent extension of StackChan by M5Stack and its community. Assistant names describe intended choices, not affiliation or a promise that every integration is available today.
Current extension work includes character UI and host-side lifecycle tests. The wider provider selection and physical companion experience are still being developed.
THE REASON TO CONNECT IT ALL
More agency. More possibility. More room to play.
Not everyone has a development team, a studio or someone to help turn an idea into a first step. We’re exploring how connected AI tools can put more of that capability within reach—and leave people free to decide what to do with it.
I bring hands-on audio, AV and IT integration into AI-assisted product development. My work connects the practical details—requirements, software, devices and testing—to a bigger picture of what people could do with them.
For a hiring conversation or project walkthrough, reply to the application or introduction that brought you here. I’d be glad to talk through the system, the choices behind it and where it could go next.