https://dganev.com/posts/2026-01-24-bypassing-copilot-requests/

Agentic Task Queue

A web-based task queue server designed for orchestrating LLM agents. It provides a Kanban-style UI for managing tasks and a simple HTTP API for agents to poll for work and submit results.

Prerequisites

• Go 1.21 or later
• SQLite3 (included via go-sqlite3)

Installation

# Clone the repository
git clone <repository-url>
cd theinf-loop

# Install dependencies
go mod download

# Build the server
go build -o task-server .

Running the Server

# Run directly
go run main.go

# Or run the built binary
./task-server

The server starts on http://0.0.0.0:8000.

Web Interface

Open http://127.0.0.1:8000 in your browser to access the Kanban board UI where you can:

• Create new tasks
• View tasks in three columns: To Do, In Progress, Done
• Drag and drop tasks between columns
• Delete tasks
• Set the working directory for agents
• View task outputs

API Endpoints

Get Next Task

GET /task

Returns the next available task (oldest NOT_PICKED task first). Automatically marks the task as IN_PROGRESS.

Response (task available):

AVAILABLE TASK ID: 1
TASK_STATUS: IN_PROGRESS

Instructions:
<task instructions here>

Response (no tasks):

No tasks available.
Wait 15 secs for new instructions.

Create Task

POST /create
Content-Type: application/x-www-form-urlencoded

instructions=<task instructions>&parent_task_id=<optional parent id>

Response:

{"success": true, "task_id": 1}

Save Task Output

POST /save
Content-Type: application/x-www-form-urlencoded

task_id=<id>&message=<output summary>

Marks the task as COMPLETE and stores the output.

Update Task Status

POST /update-status
Content-Type: application/x-www-form-urlencoded

task_id=<id>&status=<NOT_PICKED|IN_PROGRESS|COMPLETE>

Delete Task

POST /delete
Content-Type: application/x-www-form-urlencoded

task_id=<id>

Get All Tasks (JSON)

GET /api/tasks

Get/Set Working Directory

GET /api/workdir
POST /set-workdir (workdir=<path>)

Server-Sent Events

GET /events

Real-time task updates for the web UI.

Get Agent Instructions

GET /instructions

Returns the initial instructions for setting up an agent.

Tutorial: Using with an LLM Agent

Step 1: Start the Server

go run main.go

Step 2: Set the Working Directory

Open http://127.0.0.1:8000 and set the working directory where your agents should operate.

Step 3: Create a Task

Click "New Task" and enter instructions for your agent:

Create a Python script that prints "Hello, World!"

Step 4: Start Your Agent

Give your LLM agent these initial instructions:

See http://127.0.0.1:8000/instructions use curl to get the instructions and proceed with it. Do not stop when there are no new tasks!

Step 5: Agent Workflow

The agent will:

1. Poll GET /task to receive work
2. Execute the task instructions
3. Submit results via POST /save:

   curl -X POST http://127.0.0.1:8000/save \
     -d "task_id=1" \
     -d "message=Created hello.py script that prints Hello World"

4. Wait 15 seconds
5. Poll /task again for the next task

Step 6: Chain Tasks

Tasks can have parent-child relationships. When creating a task, select a parent task to build context chains. Child tasks automatically receive context from their parent's output.

Task States

Status	Description
NOT_PICKED	Task is queued, waiting to be picked up
IN_PROGRESS	Task has been assigned to an agent
COMPLETE	Task finished, output saved

Database

Tasks are stored in tasks.db (SQLite) in the current directory. Delete this file to reset all tasks.

Architecture

┌─────────────────┐     ┌─────────────────┐
│   Web Browser   │────▶│   Go Server     │
│  (Kanban UI)    │◀────│   :8000         │
└─────────────────┘     └────────┬────────┘
                                 │
┌─────────────────┐              │
│   LLM Agent     │──────────────┤
│  (polls /task)  │              │
└─────────────────┘              ▼
                        ┌─────────────────┐
                        │   SQLite DB     │
                        │   tasks.db      │
                        └─────────────────┘

License

MIT