Loop your agents on real-world outcomes.

Set the goal. Run the loop. Sharpen the strategy. Superdense gives your AI agents memory and a feedback loop tied to real outcomes — so every run compounds toward a metric that matters.

the outcome loop
$npm i -g @nimrobo/superdense
$/outcome-setup → goal.md run.md gate.md
$/outcome-run → pick a lever, test, record
$/outcome-update → promote what works, retire what fails

Meet Superdense.

Superdense is a single install that turns your agent's scattered runs into a durable, searchable feedback loop — running entirely on your machine.

Local-first

Your sessions, artifacts, and reward snapshots live in SQLite on your machine. Nothing leaves it.

Studio included

A local browser dashboard to browse sessions, compare cohorts, and surface insights — no infrastructure to stand up.

Apache-2.0

Fully open source. Read it, fork it, run it in your own loop — no lock-in.

Agent-agnostic

Adapters read sessions from Claude Code, Codex, Cursor, and OpenCode out of the box.

Claude CodeCodexCursorOpenCode
View the repo on GitHub
Coming soon

The managed outcome harness.

Run the outcome loop continuously, across goals and agents, without babysitting it. Join the waitlist to get early access.