Knowledge Hub, Per-Repo Base Branches, and a Cleaner Settings Experience
This release is built around a single question: where does your team's knowledge live -- and how does it surface exactly when someone needs it? That's what the new Knowledge Hub is for: an organization-wide place for questions, answers, and solutions, with voting, replies, and AI-assisted consolidation. Alongside that come a few smaller but tangible improvements: per-repository base branches, a consistent settings layout for Org and Project, and a much calmer experience when editing tickets.
Knowledge Hub: StackOverflow for Your Team
Spedy has had a Wiki for structured knowledge for a while, plus a Knowledge Store that agents use as their long-term memory. What was missing: a place for conversation. Ask a question, collect answers, mark the best one -- and end up with a clean entry that sticks around.
That's the Knowledge Hub.
The Flow
You open Knowledge in the sidebar and see every entry in your organization -- across all projects and boards. Filter by category, tags, or project, sort by recency or vote score. Search is hybrid: PostgreSQL full-text combined with semantic vector search.
Click into an entry and you land on a dedicated page at /knowledge/:id. The URL is shareable, the layout is full-width. The entry renders in Markdown at the top, and below it sits a flat conversation thread with replies from people and agents. Each reply supports Markdown, and each can be voted on independently.
Voting
Every entry and every reply has a net score. Click upvote or downvote, click again to undo. One vote per user. Nothing fancy -- but it's the foundation: good answers rise, weak ones fall, and the list sorts itself.
AI-Powered Consolidation
The real trick happens once a thread has become useful. Instead of leaving the conversation as a loose collection, you can hit Consolidate. Spedy uses an LLM that reads the original entry plus the entire reply thread and proposes a new, cleanly written version of the entry. The old version moves into version history, and the new state becomes version n+1.
Important: the old replies stay visible, with a visual divider marking "Consolidated into Version X". Nothing is lost, the history is traceable.
The consolidation runs in a transaction: the LLM call happens before any DB writes, so a failure leaves no half-written state behind.
Version History
Every change to an entry creates a version. You see who changed what when, can read previous versions, and restore one if needed.
Who Has Access
The Knowledge Hub is for internal roles -- admins and team members. Customers don't have access. Permissions are fine-grained:
ai-knowledge:view-- see entries and repliesai-knowledge:create-- create entries, post repliesai-knowledge:edit-- edit entries, trigger consolidationai-knowledge:delete-- delete entries or replies
The hub is part of the Pro plan and is gated behind the AI_KNOWLEDGE_BASE org feature.
How It All Fits Together
Three knowledge layers, one platform:
| Surface | What it's for |
|---|---|
| Wiki | Structured, curated documentation: specs, processes, onboarding |
| Knowledge Hub | Conversation and collective Q&A: questions, answers, versioning |
| AI Knowledge Store | The agents' operational memory: solutions, patterns, conventions |
They complement each other, they don't overlap. And all three are searchable -- the Knowledge Hub shares the hybrid search infrastructure with the AI Knowledge Store, so a single query returns hits from both sources.
Per-Repository Base Branch
Until now, the runner always used hardcoded main as the base branch. Anyone working on master, develop, or a trunk branch had to build workarounds.
From now on, every repository link on a project stores its own baseBranch. When you add a repository, Spedy automatically picks up the provider's default branch (GitHub, GitLab, Bitbucket). You can override it any time in the project configuration under Build & Validate.
What changes in practice:
- The runner branches new feature branches off the configured base, not
main - Pull requests are opened against the correct base branch
- The repository list shows a badge with the current base branch
- Multi-repo projects can use a different base branch per repository
The setup UI shows the working directory plus the base branch for each repository, so you can see at a glance where the runner will work and what it will push against.
Settings, Re-Sorted
Configuration was historically scattered across five different directories: a tab here, an inline section there, a separate area somewhere else. We consolidated it.
On the Backend
There's now a dedicated org-config module as the central entry point for all org-level endpoints: features, explainability, triggers. Previously these controllers lived in various feature modules. The Organization data model has been re-grouped with clear section comments: Core, Billing, Time Tracking, Runner, etc.
On the Frontend
All settings components -- org and project -- now live in one shared features/platform-config/ module. The project settings, which used to be a monolithic ten-tab page, are now a sidebar-based navigation -- the same pattern as org and account settings. Consistent layout, consistent behavior.
The org-settings sidebar now groups sections into four buckets:
- General -- organization data
- Configuration -- features, explainability, triggers
- Integrations & AI -- git providers, MCP, knowledge
- System -- audit, webhooks, disaster recovery
There's now a dedicated Explainability Settings page instead of hiding it as a sub-section between other config options.
Functionally nothing changes -- but if you go through settings several times a day, you'll feel the difference.
Ticket Editing: Calmer and More Stable
Several small but tangible improvements around the ticket editing experience:
Debounced Draft Patches
When you typed in a ticket draft (title, description, hours), every keystroke triggered a PATCH call. On slower connections, server responses would occasionally overwrite the word you were currently typing.
Now: 300ms debounce on text fields, local state for inputs. Server responses no longer overwrite anything while you're typing. Select fields (priority, type) stay immediate, since they're a single click.
Multiline Title
Long ticket titles used to be cut off in a narrow single-line input. Now the title field fills the full header width and wraps multi-line. Edit and copy icons align with the first line.
Inline Validation for Estimated Hours
If you enter more hours than the configured maximum allows, you now see the error directly on the input -- not only when saving.
Readable Banners
Success and error banners in the ticket chat had a color scheme that was hard to read in dark mode. We aligned them with the design system -- calmer contrasts, clear hierarchy.
Chat Features for Free-Plan Users
Free-plan users without AI tokens used to see chat buttons that threw an error on click. Those buttons are now hidden when no tokens are available. Cleaner UI, less frustration.
Other Improvements
- GitHub integration UI: Text in the method chooser cards now wraps correctly instead of overflowing. German translations on the GitHub integrations page have been completed and umlauts fixed
- Dashboard: The billing section was removed from the dashboard -- billing lives entirely under settings now, where it belongs
- Ticket description: The border around empty description fields is gone -- no more visual noise when there's nothing to show
- PR list: More resilient handling of partial data in the pull request list -- no more crashes on missing arrays
Summary
| Change | What it means |
|---|---|
| Knowledge Hub | Voting, replies, AI consolidation -- one place for team Q&A |
| Version history | Every change to a knowledge entry stays traceable |
| Per-repo base branch | No more hardcoded main, each repo can have its own default |
| Settings restructure | Consistent sidebar layout for org and project |
| Debounced drafts | Typing no longer overwritten by server responses |
| Multiline title | Long ticket titles fit, instead of being cut off |
| Inline validation | Hour errors right on the field, not just on save |
| Free-plan polish | No more dead AI buttons for users without tokens |
We're especially curious to see how the Knowledge Hub holds up day to day. Try it out, vote on what helps, and consolidate the threads you've reached for more than once.