mica

• 

  mica caption

• 

  mica demo

• 

  mica thumbnail

Inspiration

My goal was to reduce the friction of adding a production or SRE engineer to a new project. Learning an unfamiliar system while responding to real production problems takes time. Mica creates a clear path for engineers to understand the product, build operational context, and contribute sooner, with AI helping accelerate the process.

What it does

Mica is a local workspace for production investigations. It compares healthy and degraded Prometheus data, saves the evidence, and keeps hypotheses, code changes, tests, approvals, and recovery checks in one incident record. Humans use the web workspace while AI coding agents use MCP tools. Both work with the same context.

How we built it

We built the local daemon in Go, the interface in React, and the incident store with SQLite. Mica reads metrics from Prometheus and provides typed MCP tools for AI agents. We also built a Docker Compose environment with a checkout service, traffic generator, and a real N+1 regression that can be triggered, investigated, and fixed.

Challenges we ran into

The main challenge was designing one workflow that worked for both humans and AI agents without increasing cognitive load. We also had to keep every conclusion connected to real telemetry, preserve the complete investigation history, and make human approval clear for sensitive actions.

Accomplishments that we're proud of

We completed the full investigation loop: detect a regression, inspect the evidence, investigate the cause, record a change, and verify recovery with fresh telemetry. Humans and agents can continue the same incident without losing context, and the complete demo runs locally with deterministic data.

What we learned

AI agents work better when they receive structured context instead of one large prompt. Typed tools, shared state, clear evidence, and explicit boundaries make their work easier to review and trust. We also learned that production tools should reduce information overload, not add another dashboard full of noise.

What's next for Mica

Next, we want to add logs and traces, support more telemetry backends, improve multi-user collaboration, and simplify installation. We also want Mica to help new production engineers learn a system before the first incident, not only while one is already happening.

Built With

• codex
• docker
• go
• gpt-5.6-sol-high
• openai
• typescript