Umbrella Mode

Inspiration

As an engineer, I’ve spent countless hours on support calls helping users debug issues that could have been resolved automatically if the system understood the context of the problem. I wanted to build something that could think and act like a support engineer—directly inside the product.

That became the core idea behind Umbrella Mode: an AI-powered browser agent that observes what’s happening in the app, reasons about the user’s intent, and takes the right actions in real time.

What It Does

Umbrella Mode connects three key technologies:

OpenAI for reasoning and decision-making — turning plain-language user instructions into concrete browser actions.
Datadog for observability — surfacing backend logs and tracing errors to help the AI understand system state.
ClickHouse for analytics — capturing every action, validation, and error event to improve performance and agent reliability over time.

Together, these pieces create a live debugging and support experience that feels like having a 24/7 engineer inside your product.

How We Built It

The frontend form captures user intent and triggers an OpenAI-powered reasoning engine that interprets what needs to be done.
Each interaction and validation event is logged to Datadog for monitoring, while summarized data is piped into ClickHouse for querying and visualization.
The result is a transparent, observable pipeline between user intent, AI reasoning, and real execution.

Challenges

The hardest part was making Datadog searchable contextually — transforming raw logs into something an LLM could understand. Getting consistent, meaningful traces that mapped back to user actions took iteration and experimentation with both Datadog’s API and log formatting.

What We Learned

Observability data is only useful when it’s interpretable by AI.
Combining structured analytics (ClickHouse) with unstructured reasoning (OpenAI) creates powerful feedback loops.
Even a simple form can become a testbed for self-healing, intelligent systems.

Next Steps

Expand Umbrella Mode into a plug-and-play framework companies can embed into their apps to provide AI-driven in-product support powered by real-time observability and analytics.