Relmio 0.22.2

Relmio 0.22.2 makes Help usable on small and short windows and keeps the setup guide off the dashboard's buttons. It also corrects how the docs and progress messages describe ChatGPT sign-in.

Fixed

• Help now opens on short or narrow windows, such as 1280 x 320 or a phone held sideways. When the window is too short for the whole card, the card fits below the top bar and its text scrolls, with its buttons in view.
• On the local dashboard, the setup guide card and the Help chip no longer cover the selected connection's buttons, such as Review removal.
• Sign-in documentation now says ChatGPT opens in a new tab or window in the browser running Relmio when allowed, and otherwise in the default browser. Progress messages say "sign-in page" instead of "window", and a sign-in that can't start in that tab says "ChatGPT sign-in could not start. Return to Relmio and try again."
• A published source-check record no longer includes local folder paths.

Installed VPS and local sidecars need no update for this release.

Full notes: CHANGELOG.md

Relmio v0.22.1

Relmio 0.22.1 fixes Windows sign-in and owned-stack recovery, keeps Help
available without turning the walkthrough back on, and makes local image
sign-in codes easier to read and copy. Existing consent and installation
confirmations still apply.

Changed

• Contextual mascot Help stays available after skipping the walkthrough.
  It explains the prerequisite for Check the server, offers the current
  feature's tips and versioned bundled FAQ, and highlights controls without
  activating consent or installation actions.
• A new ChatGPT registration enables eligible text plan use after its first
  verified plan grant and shows a one-time Got it notice. Existing paused
  accounts stay paused; n8n background activity still needs separate approval.

Fixed

• Local image sign-in codes are larger and selectable, with a user-clicked
  Copy code button and manual-copy recovery when clipboard access is
  unavailable. Unchanged polls preserve selection; expired or replaced codes
  clear stale feedback. Copying is never automatic.
• The PowerShell installer exposes its temporary Node.js runtime on PATH
  while Relmio starts, fixing the no-Node Windows wizard launch error.
• ChatGPT sign-in opens in the browser running Relmio when a new tab is allowed,
  with a system-browser fallback and safe cancellation of an unused attempt.
  The first-plan notice keeps keyboard focus inside while open.
• Resuming an exact owned local n8n stack works without Compose's newer
  start --wait flags and verifies health before reporting success.
• Windows QA fixtures handle denied file symlinks explicitly and keep the CMD
  lock test deterministic. Generated docs retain the first line after the title.
• Escape dismisses Help during a busy operation without unlocking installation
  controls. New-n8n Help describes the private default and optional public ngrok
  link correctly.

Limits and update note

This patch does not establish fresh live VPS-update, image-generation or
SuperGrok-workflow acceptance. Those checks remain separate from the Windows
and Help evidence. The known development-tool braces advisory
GHSA-vfj7-8cjw-p6xm
remains unpatched; the owner accepted that risk with a macOS re-audit planned.
Production dependency audits are a separate check.

Wizard UI files are cached when its server starts. Browser reload alone does
not activate this UI update. Finish or cancel any current private sign-in
handoff before a coordinated wizard restart; in-memory wizard state is not
guaranteed to survive. This release preparation did not restart the live
wizard, n8n, a sidecar or production VPS services.

Full changes.

Relmio 0.22.0

Relmio 0.22.0 turns Set up new n8n into a private, one-click n8n that the ChatGPT plan, local model, SuperGrok and Assistant add-ons can join without manual wiring. It also fixes ChatGPT sign-in on Windows.

Added

• When Set up new n8n finishes, the Ready step has Open n8n and buttons to Add ChatGPT plan, Add local model, Add SuperGrok and Add Assistant tools. Each opens with this n8n and its network already selected. Add Assistant tools is hidden when you chose Code Sandbox during setup, because the stack already has it.
• For an n8n created by Relmio, an approved add-on install creates its credential in n8n: Relmio ChatGPT plan, Relmio local model or Relmio SuperGrok. Pick it in your OpenAI nodes. OpenAI tokens stay in the sidecar. If Relmio can't confirm the credential, Ready asks you to check n8n first and shows the key under Manual credential setup.

Changed

• New local n8n stacks are private to this computer by default. A public URL through ngrok is an optional setting, off by default.
• New stacks use n8n 2.42.5. Stack services and newly installed local models start again when Docker restarts. Existing local model installs keep their current setting.
• Removing a stack is refused while add-ons are still attached. Relmio names them so you can remove them first.

Fixed

• On Windows, Continue with ChatGPT opens the sign-in page in your default browser again, on this computer and for a VPS (#115). If the browser can't be started, Relmio says so right away.
• Set up new n8n with Code Sandbox or Code Sandbox + SearXNG no longer fails during setup.
• The dashboard can read a stack whose local address uses localhost.

Installed VPS and local sidecars need no update for this release.

Full notes: CHANGELOG.md

Relmio 0.21.0

Relmio 0.21.0 lets you turn on image generation while you install the ChatGPT plan sidecar, on a VPS or on this computer, so n8n's image model list is ready right after setup.

Changed

• On the install review step, open Image generation (optional), read the notes and tick the box. After a verified install, Relmio starts the Codex sign-in and the Ready step shows the image status below your one-time Relmio key. Open Enter the Codex sign-in code (VPS) or Image generation (this computer), then enter the code at https://auth.openai.com/codex/device. On a VPS, Relmio keeps the SSH connection open only until you finish or cancel, so keep the tab open.
• Leave the box unticked and the install is exactly as before. A failed image sign-in never fails the install; you can turn images on later from the sidecar's image controls.
• Before, images could only be turned on after installing, so n8n's OpenAI image Model list showed no results after a fresh install. Troubleshooting now has a row for that.
• Cancel image sign-in in the VPS image panel now also disconnects from the server.

Fixed

• At 320 pixels wide, opening Image generation on the local dashboard no longer makes the page scroll sideways.

Installed VPS sidecars need no update for this release.

Full notes: CHANGELOG.md

Relmio 0.20.0

Relmio 0.20.0 adds image generation to the local ChatGPT plan sidecar and fixes installing that sidecar for an n8n on this computer. In 0.19.0 and 0.19.1 that install always failed.

Added

• The local ChatGPT plan sidecar can generate and edit images through the same opt-in add-on as a VPS sidecar. In the local dashboard, open the installed sidecar, choose Manage ChatGPT sign-out, then Image generation and Sign in for images, and enter the code at https://auth.openai.com/codex/device. The Codex sign-in is stored in the sidecar's Docker volume on this computer, where administrators and Docker users can read it. Sign out and revoke and Remove bridge sign out of images first.

Fixed

• Installing the ChatGPT plan sidecar for an n8n on this computer works again. It failed with "Local SIWC private volume initialization failed." and then "Local SIWC installation operation failed." Both failures happened before the sign-in was handed over, so it stayed on this computer.

Changed

• The website's docs include Configure n8n, and the FAQ answers whether n8n can generate images and what the limits are.
• The images_off message now names both places to turn images on. Because the sidecar file changed, an installed VPS sidecar offers a runtime update: open the wizard, choose n8n with ChatGPT sign-in, connect and press Review sidecar update. The current VPS sidecar keeps working until you apply it.

Full notes: CHANGELOG.md

Relmio v0.19.1

Relmio 0.19.1 fixes the VPS sidecar update route and adds a ChatGPT sign-in refresh for an installed VPS sidecar. It also fixes n8n's OpenAI node and its AI Agent with Use Responses API off, keeps the wizard's star count, and adds change-aware CI.

To get the sidecar fixes on a VPS, open the wizard, choose n8n with ChatGPT sign-in, connect to your server and press Review sidecar update. Your ChatGPT sign-in and Relmio key stay the same, and n8n is not restarted.

Fixed

• On a VPS, the main button on Choose your n8n follows what is installed. Where the ChatGPT plan sidecar already runs, it reads Review sidecar update and runs the update review. Before, it led to Review bridge update, which the server refused with vps_sidecar_owned.
• n8n's OpenAI node (Message a Model) no longer fails with "Bad request - please check your parameters". The node asks for store:true by default; the sidecar now accepts it and still sends store:false to OpenAI.
• n8n's AI Agent no longer fails on its second turn when the OpenAI Chat Model has Use Responses API off.
• The local wizard's GitHub chip no longer shows ? stars when GitHub's unsigned request limit runs out. It keeps the count for 15 minutes and asks Relmio's website as a fallback. The chip still shows your running version.
• On Windows, a ChatGPT plan request no longer fails at once with "SIWC session lock is unavailable" when the lock holder exits during the check.
• relmio stop no longer fails now and then with "Relmio refuses changed local dashboard control state" while the dashboard shuts down.

Added

• Refresh ChatGPT sign-in for an installed VPS sidecar, shown only when its sign-in needs a fresh one. It signs in again on your computer with the same account and hands the new sign-in to the sidecar through the reviewed replacement. ChatGPT treats it as a new connection, and n8n needs the new Relmio key afterward.
• The setup guide explains vps_sidecar_owned, vps_sidecar_updating, vps_sidecar_staged and vps_sidecar_partial.

Changed

• Hosting options and Self-hosted model on my server start with Hostinger KVM VPS selected.
• CI runs only the checks a change can affect, adds macOS, Linux and an n8n node compatibility job (n8n 2.40.7 in Docker against the real sidecar), and reports one required CI gate check. CodeRabbit reviews every pull request.
• Release QA uses one reusable n8n workflow, "Relmio release acceptance", on any self-hosted n8n.

The 0.19.1 OpenAI source check is in docs/openai-source-check-2026-10-07.md. Release QA evidence is on #109.

Full changelog: v0.19.0...v0.19.1

v0.19.0

Relmio 0.19.0 adds a skippable setup guide with Relmio the mascot, Plan and usage counts, automatic model discovery and sidecar updates from the wizard, and moves ChatGPT plan sign-in to OpenAI's documented Sign in with ChatGPT flow.

Added

• A setup guide on every wizard page. Relmio, the mascot, points at the box or button you need next, says what it is for in plain words and shows an example where it helps. It never types into a field. Each finished step earns a badge. The first visit asks whether to start it; Hide, Skip guide and the Setup guide button in the top bar turn it off or back on. When a step fails, the guide explains the error, lists the next steps and points at the control that fixes it.
• Plan and usage, on the VPS owner panel and Ready screen and on the This computer dashboard. It shows the text requests and tokens that went through the sidecar in the last 30 days, per model, the last plan-usage error with its next step, and which models OpenAI lists and which a completed request verified. These are Relmio's own counts. Plan percent, reset times and credits stay on ChatGPT's Usage page, which Manage usage opens.
• The sidecar counts requests per day and model and stores no prompts, outputs, request IDs or IP addresses. An existing VPS sidecar starts counting after Review sidecar update.
• Automatic model discovery. The sidecar asks OpenAI's catalog as the newest stable Codex release (checked on npm about every 12 hours), so new text models appear without a Relmio release. VPS installs can opt in to short model checks, and Check installed account shows each model's state with a Copy ID button.
• Review sidecar update, which rebuilds and restarts only the VPS sidecar from the current Relmio version. The ChatGPT sign-in and the Relmio key stay the same, and n8n is not touched.
• An opt-in image add-on for a VPS sidecar (gpt-image-2 through the OpenAI node's image actions), using a separate Codex sign-in. OpenAI does not document this route for other apps, so it can stop working without notice.
• Chat Completions tool calls, staged-install resume and a reviewed reconcile when a session handoff's acknowledgment was lost.

Changed

• ChatGPT plan use now goes through OpenAI's documented Sign in with ChatGPT flow, with per-registration credentials kept local and documented transfer boundaries for n8n and a VPS.
• Start cards say who each route is for and what you need. The step list is a compact quest track, and Ready screens list the badges earned and the next steps. The setup flow itself is unchanged.
• n8n's AI Assistant can use the sidecar through its Self-hosted or OpenAI-compatible endpoint. On 2026-10-07 it worked live on one Pro account with gpt-6-astra. Only do this if nobody else uses that n8n; whether it fits OpenAI's Sign in with ChatGPT terms is open.
• The dark theme's accent is pastel green.

Fixed

• Sign-in no longer fails on OpenAI's list-form ID token audience or its optional callback scope, and declining shows a declined message.
• Non-streaming requests no longer fail when OpenAI's final event has an empty output, and streamed failures reach AI SDK clients as errors.
• n8n's AI Assistant model check is no longer refused: output-token caps that ChatGPT plan use cannot honor are dropped instead of refused.
• VPS installs and resumes no longer stop after the image build, and a resume rebuilds an outdated sidecar image.
• Session locks survive crashes, restarts and Windows antivirus holds, and Windows owner-only checks reuse one PowerShell process.

Security

• VPS installs bind the reviewed n8n container and network and check them again before writing. SIWC files are published atomically under /docker/n8n-openai-oauth, and every SSH command has a finite deadline.
• Recovery never restarts an old writer automatically.

Known

• The hosted site's full npm audit reports high advisories in dev-only build and lint tools (no patched braces release). The production dependency audit is clean.
• OpenAI source checks for this release are in docs/openai-source-check-2026-10-05.md, -2026-10-06.md and -2026-10-07.md, including open Terms questions.

Full Changelog: v0.18.6...v0.19.0

Relmio v0.18.6

Relmio 0.18.6 makes the selected theme and keyboard focus easier to distinguish.

Fixed

• The selected theme in the local wizard and website has a high-contrast inset ring in both themes, so it stays visible after focus moves away. In forced colors, selection and keyboard focus have separate indicators. Theme behavior and control sizes are unchanged.

Full Changelog: v0.18.5...v0.18.6

Relmio v0.18.5

Relmio 0.18.5 changes the light theme from pastel yellow to pastel green.

Changed

• The light theme of the local wizard and the website is now pastel green instead of pastel yellow. Pages, panels, lines, selected items and the home page's day scene use the new greens, and success messages use a cooler green so they stay distinct from the page. Text contrast still meets WCAG AA. The dark theme is unchanged.

Relmio v0.18.4

Relmio 0.18.4 restores browser format checks on the /hosting plan details.

Fixed

• Hosting plan detail fields on /hosting check their format in the browser again. Most field patterns used characters that browsers reject, so the browser skipped the check and a malformed value only failed later with a general error. Accepted values are unchanged.