upcontrol

Monitoring that arrives with the "why", not just "down" - wired into your app by the coding agent you already use.

npx upcontrol

One command, for every coding agent - the badge above links to the current list. It does four deterministic things and runs no AI of its own:

1. Installs the upcontrol skill into .claude/skills/ and .agents/skills/ (add --copilot for .github/skills/) - event-name guidance, placement rules and stack recipes your agent follows.
2. Pins @upcontrol/sdk in package.json at an exact version - the library whose track() never throws and never blocks.
3. Provisions a key. No account needed: an anonymous project is minted and the key is written into .env - only after .gitignore is confirmed to cover it, and without ever printing it. Claim the project later (free, same key) via the printed link. Already have a key? --key uc_live_... or UPCONTROL_API_KEY.
4. Verifies. npx upcontrol verify waits until data provably arrives and names the failure precisely when it does not.

Then talk to your agent in plain language:

send all my logs to upcontrol

track user behavior and tell me where customers churn

alert me when my nightly cron dies silently

The agent reads the skill, stages a diff for your review (it never commits), and finishes only when verify reports data flowing.

Run inside an agent, the command detects it and answers in JSON - so "set up upcontrol for me" works as a single prompt too.

• npx upcontrol skills - the reference topics the agent reads
• npx upcontrol status - endpoint, key source, skill freshness, one JSON line
• npx upcontrol board - read, replace (--apply) or extend (--add) the project's dashboard as JSON, so the agent can build or grow it from the events it just wired up. A project can hold several named boards: --list shows them, --new <name> makes one and --board <id|name> picks which one a read or a write means
• Your code never leaves your machine: the CLI talks only to the upcontrol API, the intelligence is your own agent, and the SDK sends only what the reviewed log points emit - scrubbed client-side first. What leaves is the log lines your own code emits, and nothing on our side ever feeds them to a language model.

https://upcontrol.io