Open public beta

One GPU. Real machine.

DSH-RSI

A coding agent running on a GPU in someone's apartment.

Qwen 3.8 27B from a single RTX 5090. Sessions pinned to an immutable runtime release—the harness that starts a task is the harness that finishes it.

01 / IdentityEmail verified
02 / ComputeSingle RTX 5090
03 / RuntimeImmutable release

What is this

Your agent. It grows.

A personal coding agent that keeps what it learns. It remembers your projects across sessions, builds itself new tools, and can rewrite parts of how it works — everything it becomes stays on your private volume. Nobody else's agent learns from yours; ours don't reset yours.

How it works

Four honest parts.

01 / IsolationYour own machine-in-a-machine

An isolated container with a private encrypted volume. Read-only system, resource caps, locked-down network. Snapshotted daily, 7-day retention.

02 / Real toolsIt actually ships

Runs bash, edits files, clones from GitHub/GitLab, installs from npm / PyPI / crates through an allowlisted proxy. Upload files in; search the live web out.

03 / The modelOne GPU, no fine print

Qwen 3.8 27B served from a single RTX 5090 over a private link. All tenants share one FIFO queue — honest latency, no overselling.

04 / EvolutionImprovements persist

The agent can extend itself in memory, then promote what works into its permanent runtime. Bad promotions auto-revert; a broken profile self-quarantines. You can't brick it.

Honest limits

Read before you lean on it.

Create account Read the docs