Skip to content

roam v0.1.0 — leave a working agent on a remote box

Choose a tag to compare

View all tags
@javimosch javimosch released this 10 Jul 10:26
· 46 commits to main since this release
v0.1.0
3837797

First release. Dispatch a self-replicating autonomous agent to a VM, walk away, and attach back to steer it — one static binary, no Python on the target.

Install

curl -fsSL https://github.com/javimosch/roam/releases/latest/download/roam-x86_64-linux -o roam
chmod +x roam && ./roam help

roam-x86_64-linux is fully static (bundled SQLite + static OpenSSL + CA store) — runs FROM scratch on any x86-64 Linux, no deps. Verify with roam-x86_64-linux.sha256.

Highlights

  • Remote dispatch + self-replication over ssh; control (status/attach/steer/stop) against the remote journal.
  • Real LLM tool-loop (read/write/shell/finish), written in machin — no SDK. Providers: Anthropic and OpenAI-compatible (OpenRouter, …).
  • Trust layer: hard token/iteration budgets, workdir sandbox, gated shell, full journal.
  • Confirm-gate — destructive commands park for async approval (approve/deny/stop), with a deny budget. Live-proven (caught a sudo rm escalation).
  • Goal-verify — an independent judge checks completion against real evidence; fail-open.

Docs + changelog: https://javimosch.github.io/roam/ · Full notes: CHANGELOG.md

Assets 4
Loading