Preview branch selection and grouped notifications

Five areas in this update: previews now boot any branch or pull request instead of just the default branch, the notification panel groups by category, the Issues Hub gains a due date column, AI agent settings move into their own sidebar group with a provider status card, and registration now runs with manual approval.

Previews now boot any branch

Until now, a preview always cloned the repo's default branch on start. When the AI agent had been working on a feature branch and opened a pull request, you couldn't see exactly that state as a running application — even though that's the code you actually want to review.

Now you can: the preview's start card has a searchable branch dropdown. It lists the repo's default branch plus every branch and pull request linked to the ticket. Pick the branch behind an open PR, click Start preview — and see exactly the code under review running.

Two details worth knowing:

• If you don't pick anything, the default branch boots — the existing behavior is unchanged.
• The Reset button rebuilds the preview on the branch it was started with — no longer stubbornly on the default branch.

The agent writes the code, you watch it run before anything gets merged — the decision stays with you.

Notifications grouped by category

The notification panel used to be one flat, chronological list. On a busy board with automations and runner jobs, that got noisy fast: budget warnings, ticket assignments, and automation messages all mixed together.

The panel now groups into five categories, each with a header that stays visible while you scroll:

• Tickets — assignments, mentions, status changes, comments
• Budget — threshold warnings and overruns
• Automations — messages from automation and runner runs
• Personal Todos — reminders for your todos
• Workspace — invitations and organization-wide messages

Within each category, entries remain chronological. Filters (All / Unread) and “Mark all as read” work as before. More in the notifications docs.

Due Date column in the Issues Hub

The Issues Hub can now show a Due Date column with each ticket's due date. The column is hidden by default — enable it via the list view's column configurator. Click the column header to sort ascending or descending, and you'll instantly see what's due next across all your boards.

A dedicated home for AI agent settings

The settings sidebar used to mix integrations and AI configuration in one “Integrations & AI” group. That's now cleanly separated:

• Integrations keeps GitHub, GitLab, webhooks, and friends.
• The new AI Agents group bundles Runner & Agents, MCP Servers, and Agent Skills.

Two labels got sharpened along the way: the former “Runners” entry is now called Runner & Agents — the page has long covered more than runner management (jobs, agent pipelines, settings, secrets, analytics). And the board-level “Agents” tab is now called Agent Setup, so it can't be confused with the org-level Agents concept.

Also new on the page: the AI provider status card. It shows at a glance whether an API key is stored in the vault for Claude (Anthropic) and OpenAI (Codex) — with a direct link to Secrets. One thing to keep in mind: runners logged in via subscription (claude /login or codex login directly on the runner) don't show up here — you still check that on the runner itself.

OpenAI model selection: consolidated to GPT-5.3 Codex

The model picker for OpenAI-based runner jobs got a cleanup. Instead of three entries (GPT-5 Codex, GPT-5, GPT-5 Mini), there's now exactly one verified working model: GPT-5.3 Codex.

The background: OpenAI's Codex model lineup churns frequently, and individual model ids get rejected depending on how the runner is logged in (API key vs. ChatGPT subscription). So we now only list what reliably works in both modes.

What you need to do: if a pipeline stage or agent still has one of the old GPT models configured, switch it to GPT-5.3 Codex. The Claude model list is not affected. Full details in the runner docs.

Registration with manual approval

We're currently opening Spedy up gradually. New registrations therefore go through manual approval from now on:

1. You sign up as usual — via the form or social login. A notice on the page explains the process.
2. After submitting, Spedy confirms: “Almost there — we're reviewing your sign-up.”
3. We review every registration and unlock access. As soon as you're in, we'll let you know by email.

Until approval, login for the new organization is blocked — with a clear message instead of a cryptic error. Nothing changes for existing organizations. Details under Create account.

Everything is available now. Questions or feedback? Reach out via the contact form.