Agent Observability
A real-time log panel shows every tool call and decision the agent makes. The operator always stays in control with full visibility into what's happening.
featured project
GHCP Agent UI
A custom interface by JT Broad for orchestrating autonomous AI agents with the GitHub Copilot SDK — built for visibility, control, and iteration.
Overview
Most AI coding tools give you a single prompt box and hope for the best. GHCP Agent UI takes a different approach: it lets a team run an AI agent in a continuous autonomous loop — pulling tasks from an inbox-style queue, executing multi-step work, and streaming real-time feedback through a chat interface. The operator stays in control with start/stop mechanics, session history, and full visibility into every decision the agent makes.
Built on the GitHub Copilot SDK, the system connects directly to Copilot's language models while layering on custom tool orchestration, persistent memory, and channel integrations — turning a single-turn assistant into a long-running autonomous worker.
Key Features
⟳
Agent Loop Execution
Start and stop autonomous task processing. The agent continuously pulls work from the queue, executes it, and moves on — no manual prompting required.
▦
Inbox Queue Management
A triage-friendly inbox keeps active work visible, easy to reprioritize, and ready for both manual review and autonomous execution.
◈
Real-Time Streaming Chat
Watch the agent think in real time via streaming responses from the Copilot SDK. Full chat history with markdown rendering.
⊞
Session Management
Persistent session history lets you review past agent runs, replay decisions, and resume work across sessions.
⇌
Channel Connectors
Plug in Telegram, Discord, or other messaging platforms to dispatch tasks and receive agent updates from anywhere.
Under the Hood
Inbox-Style Workflow
Tasks flow through a Gmail-inspired inbox view with keyboard navigation, resizable panels, and focus management for high-volume workflows.
MCP Server Integration
Connect and configure external MCP servers to extend agent capabilities with custom tools and environment variables.
Custom Agent Definitions
Assign different agent configurations per task, with skill paths and custom instructions loaded from the filesystem.
Tech Stack
- Astro 5
- React 19
- GitHub Copilot SDK
- Tailwind CSS
- shadcn/ui
- Monaco Editor
- TypeScript
Screenshots
Roadmap
The project is already usable end to end, with the remaining work focused on memory depth, external channels, and turning generated artifacts into directly served app surfaces.
-
Chat UI using Copilot SDK under the hood
-
Inbox-style issue triage with an agent queue and user inbox
-
Agent loop for processing agent queue and moving issues between categories
-
Scheduled tasks using same issue format, for a set date or recurring
-
Drive area for visualizing where artifact files are stored
-
Connect chat to running issues so they communicate back to the chat when finished
-
Connect MCP servers
-
Connect skills and custom agent definitions via file path location
-
Get alpha build to early users for feedback
-
Simple Zettelkasten-style memory system
-
Channels to connect through Discord, Telegram, or other messaging platforms
-
Apps section for extensible apps created from artifacts and immediately served