Marketing Tool Stack by Amit Gupta
Design Studio
Mission Control

A dashboard for your AI agents

Vibe coding is easy. Knowing whether your agents actually ran, what they cost, and which tools failed is not. Mission Control reads the logs your agent runtime already writes and shows you the truth. Free, self-hostable, private by design.

Node.js 18+ · deploy to Vercel in minutes · numbers below are sample data
Mission Control overview showing agents, runs, spend, and a commit chart on sample data

What it shows you

Four questions, answered from the logs instead of from memory.

Who actually ran

Real invocations per agent, not name-drops. The agent sitting at zero is either dead weight or a gap in your process.

What it cost

Exact token spend by model and by day, the main thread versus sub-agent split, and the money your cache saved.

Which tools fail

Every tool call ranked, with the error share overlaid. High volume plus high errors is where your retries hide.

How long work took

Every session with its length, turn count, agents spawned, tokens, and cost. Spot the run that ran long.

Workload view with five active agents and one idle reviewer agent at zero runs
Real runs per agent. The idle one is a decision waiting to be made. Sample data.

Deploy your own in four steps

The whole thing is an afternoon. It reads what your runtime already logs. No SDK, no instrumentation.

  1. Download and unzip the app, then put it in a private git repo.
    npm run dev # preview locally at http://127.0.0.1:4400
  2. Create a Vercel project from that repo. If it is a subfolder, set the Root Directory to it.
  3. Set one environment variable, DASH_PASSWORD, to your access password, then deploy.
  4. Generate and publish your data.
    npm run push # snapshot your ~/.claude logs, commit, deploy
Private by design. The hosted app is gated behind your password, marked no-index so it never shows up in search, and reachable only by direct link. It serves your data only through the gated endpoint, so even the raw file cannot be reached without signing in. Keep the repo private.

Stop trusting output. Start reading the logs.

Anyone can wire up a team of agents. Knowing whether they pulled their weight is the part worth building.