Roastwarden

Releases ·

Roastwarden 0.30.0

roastwarden.com now speaks to roasters: "Understand every roast. Build on every batch.", docs that open with what a roaster needs (the design notes stay in the repository), and an AI agent set up in three commands, roast login, roast skill install and roast status. On a phone the Roast page's chart is about three quarters wider.

Changed

  • roastwarden.com speaks to roasters: the home page leads with what the app does for a roaster, "Understand every roast. Build on every batch.", under "Self-hosted coffee roasting software, written in Rust"; four outcomes (see what's happening, learn what changed, repeat what worked, keep the batch together) replace the architecture pillars, and "The roast runs on the server" explains the browser is only a window onto the roast. "Will it work with my roaster?" answers logging and control apart. "Any roaster, safely" became "Connect your roaster, choose the control", which says what it does.

  • The docs have a roaster's part and design notes: each published doc opens with what a roaster needs (what it is, how to use it, its settings, the roast command), and the site publishes only that part; the design notes stay below a <!-- site: end --> line in the same file, so the two cannot drift apart. The build fails if the published part names another project, a roadmap item, a dated decision or a code path (site/README.md, "Each doc has two readers").

  • The Roast page's chart on a phone gives the curves the room: below 560 px wide the temperature and RoR scales are narrower and carry no unit labels, and the 0–100 % scale is left out (heater and fan show in the readouts and on touch), so the plot is about three quarters wider. Wider screens are unchanged.

  • The simulated roaster runs the fluid bed's tuned PI gains (8.355 / 0.22269 / 0, from its server autotune) instead of the gains from before its tuning, whose D term made the heater jump with every noisy reading: the simulator's heater now looks like the real roaster's.

  • The site's and README's pictures: the Roast page (laptop and phone) retaken with a smooth heater, Label Studio showing the built-in designs, and Reports in the home page's gallery.

Added

  • An agent set up in three commands (Agents, "Setting up an agent"): roast login <server> checks an API token and keeps the server and token in ~/.config/roastwarden/roast.json (mode 600; never the owner key), so roast and an agent's shell need no environment; roast skill install claude-code | hermes | agents (or --dir) writes the Agent Skill, now built into roast, where Claude Code, Hermes or Codex and other Agent Skills readers find it, naming the server; roast skill show prints it for any other agent; roast logout forgets the token. Settings → Tokens shows the three lines with the server's address once a token is made. A flag or ROASTWARDEN_URL / ROASTWARDEN_TOKEN still wins over what was saved.
  • The home page shows what an agent does: "Ask your agent", with what you can say and how its token, approvals and the audit log keep it in hand; Reports, typeset by the built-in Typst engine, is a feature of its own.
  • Search: the home page's title and description name what it is ("self-hosted coffee roasting software", Artisan Roaster Scope in full), schema.org SoftwareApplication data, and each release tells IndexNow (Bing and others) about every page.

Fixed

  • The Read-only token preset's description said it reads the audit log; only Admin tokens and the owner key do. roast audit --help now lists the labels kind.