A Mindjax Venture Lab experiment
Mindjax Command Center
An experimental command center for coordinating AI agents, workflows, tools, and business experiments from a single interface.
AI / INTERNAL TOOLS
Problem
Useful work gets scattered across chats, coding agents, research tools, and external services. I wanted to explore what happens when those capabilities become parts of one working system. Command Center starts with a concrete question: how does a research agent's output become something I can review, track, and act on?
Approach
Start with a workflow that can be run and inspected. Use AI-assisted development to shorten the build-and-feedback cycle, then connect the interface, tools, data, and prompts around that workflow. The first integration is deliberately narrow: run Scout, inspect the evidence, and decide what deserves follow-up. The agent supplies candidates; a person decides what enters the pipeline.
Build
The application uses Next.js App Router, React, and TypeScript with custom CSS. Server-side code launches Scout as a configured local process, captures run status and output, and parses its Markdown reports. SQLite stores runs, candidates, opportunities, discovery interviews, and activity history. API routes handle Scout launch and status; server actions handle review and record updates. This is a local Node application with filesystem access and a writable database, not a static site.
Implemented
- Scout launch and run-status tracking through API routes, with a guard against concurrent runs.
- Markdown report import into a candidate review queue; explicit Add to pipeline and Ignore actions.
- Opportunity records with evidence, source context, status, notes, and deletion support.
- Discovery-interview records, activity history, and SQLite migrations.
- Pipeline context passed to the real Scout process, plus an included demo runner for local testing.
Experimental / planned
- Coordinating more agents and tools beyond the current Scout workflow.
- Evaluating which research signals lead to useful customer-discovery work.
- Shared or hosted use: authentication and multi-user access are not implemented in this version.
Technologies
- Next.js App Router
- React
- TypeScript
- CSS
- Node.js process APIs
- SQLite / better-sqlite3
Outcome
The current build is a working local environment for testing the path from research output to human review and follow-up. It provides persistent state and an inspectable run history around the agent. That is a useful foundation for further Mindjax experiments, while broader orchestration remains work to explore rather than a finished product.
Screenshots
What I learned
- A capable model is not a workflow. It needs tools, state, and a place for its output to go.
- Context has to cross the boundary between tools. Passing pipeline context is part of the integration, not just prompt wording.
- Review is a product feature. Imported results stay separate from the pipeline until a person accepts them.
- Working prototypes reveal the unglamorous requirements: run status, parsing, persistence, and failure handling.