I got tired of drowning in terminal windows and manual git commits while vibecoding, so I built The Nexus: an open-source cyberpunk local command center to manage, orchestrate, and automate my AI projects.
My background is in Operations Research, not coding. I love designing complicated workflows, but really hated staring at code all day. When GenAI arrived, my workspace quickly got chaotic with different scripts and agents. I wanted to apply my operations logic to AI orchestration. What started as a visual dashboard quickly became a project of using agents to automate the development process itself.
The Nexus bridges your local filesystem with a web dashboard. It’s built on Next.js 16, Python, FastAPI, LangGraph, and SQLite.
The Core Experience:
- The Dashboard: A unified grid showing active projects and git status. A single click calls a Committing Agent to review codebase changes, draft a message, and push to GitHub.
- Project Workspace: Project Task Management keeps your projects organized and planned ahead. A Documentation Agent manages full project context so you don't waste tokens re-explaining things to fresh instances of Claude.
- Nexus Prime Workflow: An 8-node LangGraph orchestrator delegates tasks into four phases: Research, Plan, Implementation, and Testing.
- Human-in-the-Loop: No "black boxes." You see agent thoughts, tool calls, and real-time status. Agents propose files and terminal commands, but nothing writes to disk without GUI approval.
The Cooperative Mesh (Council of 4):
My favorite feature is the New Project Process. It kicks off in the terminal and uses a mesh of agents:
1. The Proposer: A randomizer selects Gemini Pro or Claude Opus to draft a proposal on a shared Blackboard.
2. The Council of 4: Four specialized agents (different models) provide line-level debates and suggest improvements based on their unique viewpoints.
3. Consensus: Votes are submitted and the plan is updated. It only reaches my desk once all AI specialists approve.
4. Execution: After my final human review, the system creates a local folder, initiates a git repo, creates tasks, assigns workflows, and we're ready to get to work.
Finishing Touches:
- SOTA Model Discovery: On startup, the server pings the APIs of the 4 main AI providers to auto-detect and use the latest model versions.
- System Monitor: Tracks active ports, CPU, and detailed API token usage/costs natively.
- Auto Research: Suggests new features based on project context and creates actionable tasks.
We are still in the early stages (lots of bugs to squash, and the Workflow Builder is mostly still a viewer), but The Nexus is now one of my primary workspaces and I’m actively using it to build itself.
I’d love your feedback on the architecture, the LangGraph routing logic, or the Council review concept!
Hi HN,
I got tired of drowning in terminal windows and manual git commits while vibecoding, so I built The Nexus: an open-source cyberpunk local command center to manage, orchestrate, and automate my AI projects.
My background is in Operations Research, not coding. I love designing complicated workflows, but really hated staring at code all day. When GenAI arrived, my workspace quickly got chaotic with different scripts and agents. I wanted to apply my operations logic to AI orchestration. What started as a visual dashboard quickly became a project of using agents to automate the development process itself.
The Nexus bridges your local filesystem with a web dashboard. It’s built on Next.js 16, Python, FastAPI, LangGraph, and SQLite.
The Core Experience: - The Dashboard: A unified grid showing active projects and git status. A single click calls a Committing Agent to review codebase changes, draft a message, and push to GitHub. - Project Workspace: Project Task Management keeps your projects organized and planned ahead. A Documentation Agent manages full project context so you don't waste tokens re-explaining things to fresh instances of Claude. - Nexus Prime Workflow: An 8-node LangGraph orchestrator delegates tasks into four phases: Research, Plan, Implementation, and Testing. - Human-in-the-Loop: No "black boxes." You see agent thoughts, tool calls, and real-time status. Agents propose files and terminal commands, but nothing writes to disk without GUI approval.
The Cooperative Mesh (Council of 4): My favorite feature is the New Project Process. It kicks off in the terminal and uses a mesh of agents: 1. The Proposer: A randomizer selects Gemini Pro or Claude Opus to draft a proposal on a shared Blackboard. 2. The Council of 4: Four specialized agents (different models) provide line-level debates and suggest improvements based on their unique viewpoints. 3. Consensus: Votes are submitted and the plan is updated. It only reaches my desk once all AI specialists approve. 4. Execution: After my final human review, the system creates a local folder, initiates a git repo, creates tasks, assigns workflows, and we're ready to get to work.
Finishing Touches: - SOTA Model Discovery: On startup, the server pings the APIs of the 4 main AI providers to auto-detect and use the latest model versions. - System Monitor: Tracks active ports, CPU, and detailed API token usage/costs natively. - Auto Research: Suggests new features based on project context and creates actionable tasks.
We are still in the early stages (lots of bugs to squash, and the Workflow Builder is mostly still a viewer), but The Nexus is now one of my primary workspaces and I’m actively using it to build itself.
I’d love your feedback on the architecture, the LangGraph routing logic, or the Council review concept!