Web platform
I built the Laravel Hub, React client, visual editors, apps, toolkits, Knowledge Base, authentication, billing, and community features.
Founder and sole developer
Elliot Mousseau
I have spent more than 20 years building software. My work has covered games, web and mobile apps, VR, machine learning, defense simulation, and the systems that hold all of those things together.
Over the last twelve months I built every surface of FlowDot myself. During the same period I led junior engineers and helped deliver a Phase II SBIR defense training project that landed successfully.
Elliot Mousseau
Peru, New York | FLOWDOT LLC
FlowDot brings together systems I have worked with throughout my career.I wrote the software for every surface and I still work directly in every part of the platform.
About me
I have never been interested in staying inside one small part of a system. If a problem crosses from a web application into a game client or from a cloud service into a piece of hardware, I follow it across that boundary. I want to understand the whole problem before I decide what the solution should look like.
I think empathy is a practical engineering skill. When something does not make sense, I try to understand what the person or system was trying to do and what constraints it was working under. That habit has helped me work through old software, unfamiliar systems, and products where the real problem was not sitting in the first place anyone thought to look.
Before FlowDot I built and operated the Super Contraption game series, which reached more than three million players. I also built mobile apps, government software, VR systems, and a machine learning gesture detector that recognized more than 50 gestures. At Obsidian Solutions Group I worked across the VERTEX defense training system while leading a three-person development team.
I am most useful when the problem is difficult, crosses several systems, and still has to ship.
The last twelve months
I built every surface of FlowDot.
This is a working platform, not a product mockup. I built the public site, the applications people use, the local runtimes, and the server systems behind them.
Agent runtime
I built recipes, workflows, agent conversations, goals, loops, model routing, streaming, and human approval gates.
CLI and daemon
I built the local agent, its file and code tools, persistent conversations, recurring work, and state shared across surfaces.
Desktop, mobile, and voice
I built the Electron desktop app, mobile app, live voice agents, local MCP hosting, and cross-device session control.
MCP in both directions
FlowDot can be used from other MCP clients. The desktop app can also run third-party MCP servers and give their tools to a FlowDot agent.
Permissions and audit
I built the permission system, signed audit trail, real-time run visibility, panic stop, and local model routes.
AI is the product
AI-native. Model-independent.
FlowDot needs AI to do its job. Agents use models to reason, work with tools, run recipes, search a knowledge base, remember context, and hold voice conversations. Those are not optional extras. They are central parts of the product.
FlowDot is not tied to one model company. When models get better, every FlowDot surface can use those improvements. The rest of the platform keeps the tools, permissions, run history, and user context in place when a model changes.
ModelsCloud models, local models, BYOK providers, and subscription-backed models
ProcessesAgents, recipes, workflows, branches, loops, parallel work, and gates
ContextKnowledge Base, memories, conversations, files, and page state
ToolsToolkits, MCP, workflows, local code, apps, and communications
ControlPermissions, signed audit, panic stop, run history, and human approval
SurfacesWeb, desktop, mobile, CLI, voice, VR, IDE, and comms relay
Work before FlowDot
I have shipped software under real constraints.
Some of my work has been for millions of game players. Some of it has run inside AWS GovCloud for military training. In both cases the software had to work for the people using it.
VERTEX | Defense training
Phase II SBIR delivery
I led junior engineers and worked across the VERTEX military firefighter training system. The Phase II SBIR project landed successfully.
- Unity 6 on WebGL, Meta Quest VR, and PC
- Node.js multiplayer over WebSocket and UDP
- AWS GovCloud and AI services
- DIS/HLA military simulation interoperability
- Technical direction and code review for three developers
InventaPlay | Independent games
More than three million players
I built the Super Contraption games as an independent developer. That included the games, servers, art, animation, and the systems used to share player creations.
- More than 3 million players
- User-created content and sharing
- Android and iOS releases
- A 50+ gesture ML system built in under two months
- VR releases for Quest and SteamVR
Work history
I have been building products for a long time.
2014
Super Contraption released
I released a physics sandbox with a Rails backend and a system for sharing player creations. The series later reached more than three million players.
2016 to present
InventaPlay
I built and operated games, VR projects, backend services, 3D assets, animation, and machine learning systems.
2019 to 2024
Bluehouse Group
I built iOS and Android products, led the VMBA apps, rebuilt PATH Companion with HealthKit, worked on older web systems, and mentored engineers.
2024 to 2026
VERTEX and the Phase II SBIR
I worked across Unity, multiplayer networking, Laravel, GovCloud, IoT, and military simulation standards while leading three developers.
Last 12 months
FlowDot
I built the web platform, CLI, desktop app, mobile app, voice system, MCP tools, local daemon, and the server systems behind them.
Dec. 15, 2025
FLOWDOT LLC formed in New York
I formed the company in New York and continue to build the platform from Peru, New York.
The part that is hard to copy
FlowDot is one connected system.
I am not claiming that FlowDot owns a secret model. The hard part is making agents useful across local and cloud software while keeping their work visible and under the user's control.
Where is the moat?
In the layer around the model. Permission gates before every consequential action, a signed audit trail behind it, state shared across every surface, and an isolation model where one account's data cannot be reached from another and my own properties cannot reach each other without an explicit grant. That is slow and difficult to build, and it does not get easier when models get better.
I do not claim a proprietary data advantage or a proven distribution advantage today. I would rather state that plainly than invent one for a pitch.
What happens when models improve?
FlowDot gets better with them. A better model can be used by the agent, recipes, workflows, voice system, and every other surface without rebuilding the platform around a new vendor.
01
State across surfaces
A job started on one surface can be seen and controlled from another.
02
Parts that work together
Workflows, recipes, tools, apps, MCP, the Knowledge Base, memories, goals, and loops share the same platform.
03
User control
Permissions, signed audit records, run history, remote approvals, and the panic stop are built into the system.
04
Model choice
Users decide which models handle simple, capable, and complex work. They can change that mapping later.
05
One architecture
I built the layers together, so changes in one part account for the rest of the platform.
See FlowDot for yourself.
The product is working today. You can read about the platform and inspect the public parts of it directly.