BurnWatch

Inspiration

AI coding agents can rapidly consume tokens through long-running tasks, repeated tool calls, and unattended execution flows. As agent workflows become more autonomous, it becomes increasingly difficult to understand how quickly usage and cost can grow in real time.

We wanted to build a lightweight monitoring layer that gives teams immediate visibility into AI agent token burn before costs become difficult to control.

What it does

BurnWatch is a lightweight monitoring dashboard for AI coding agents.

It tracks token usage in real time, compares usage against a configurable token budget, and visualizes risk levels through Safe, Warning, and Critical states.

The dashboard supports:

Live usage monitoring
Manual and automatic refresh modes
Budget threshold tracking
Kill Switch interaction
External API-based usage integration

BurnWatch is designed to work with CLōD-compatible AI execution pipelines and lightweight agent workflows.

How we built it

We built BurnWatch as a lightweight frontend-first monitoring dashboard using:

Next.js
React
TypeScript
Tailwind CSS
shadcn/ui

The frontend continuously fetches token usage data from an external usage endpoint and calculates budget thresholds locally in the dashboard.

We also integrated:

Auto-refresh polling
Kill Switch behavior
API response validation
Lightweight usage monitoring logic for CLōD-compatible workflows

The architecture was intentionally kept simple and modular to support fast iteration during development.

Challenges we ran into

One of the biggest challenges was defining a clean integration flow between the frontend dashboard and external AI execution systems.

Because token usage data can come from different providers and execution pipelines, we had to design the dashboard around a flexible API contract instead of tightly coupling it to a single backend implementation.

We also had to balance real-time updates with UI stability, avoiding excessive polling and layout shifts while keeping the dashboard responsive.

Accomplishments that we're proud of

We're proud that we were able to turn a complex AI infrastructure problem into a lightweight and understandable monitoring experience.

Some highlights include:

Real-time usage tracking
Configurable budget monitoring
Auto-refresh polling
Kill Switch interaction
CLōD-compatible usage integration flow
Clean and responsive dashboard UX

We also designed the system to remain simple enough for rapid iteration while still supporting future backend expansion.

What we learned

We learned that even simple AI monitoring workflows quickly become complex once real-time execution, usage tracking, and external integrations are involved.

We also learned the importance of keeping system contracts lightweight and flexible, especially when multiple teammates are working on different parts of the stack simultaneously.

Building BurnWatch reinforced how important observability and cost visibility are becoming in AI agent workflows.