Research Workspace
The Research Workspace is a dedicated space for clarifying what you want to build before writing any code. Instead of jumping straight into the App Builder and then pivoting repeatedly as requirements become clearer, you use Research to have a structured conversation with an AI analyst that helps you:
- Articulate the problem and who it affects
- Define what data needs to be stored, queried, or integrated
- Surface open questions, edge cases, and constraints
- Produce a concise project brief that seeds your new app with full context
When you're ready to build, one click graduates your session into a real Vulcan project pre-loaded with everything the AI learned.
Live at: vulcan.shipveho.com/research
How to get started
- Open the Research Workspace — three ways to get there:
- Click Research first on the Vulcan home page
- Click Research in the nav bar from the Projects page
- Go directly to vulcan.shipveho.com/research
- Enter a session title (e.g. "Driver app redesign" or "Ops dashboard for parcel exceptions") and click Start Research.
- Describe the problem you're trying to solve in the chat. The AI will ask clarifying questions.
- Keep the conversation going until you feel like the requirements are clear.
- Click Start Building to graduate the session to an App Builder project.
The Chat Interface
The Research chat works like a conversation with a business analyst. The AI persona — Vulcan Research — is tuned to ask why before how, and will push back on vague requirements rather than just accepting them.
How responses are delivered
Research chat runs on Cloudflare Workflows — a durable background job that runs independently of your browser connection. This means:
- Closing the tab is safe — the AI will finish processing and the response will be waiting when you return.
- Cancelling is not supported — once a message is submitted, the AI runs to completion regardless of whether you stay on the page. If you need to stop it, just send a follow-up message redirecting the conversation.
- Complex queries don't time out — the AI has up to 15 minutes per turn to fetch URLs, analyze documents, and save findings. There is no 30-second hard cutoff.
- Up to ~1 second delay before first token — the SSE relay polls for new output every second, so you may notice a brief pause before streaming begins.
What the AI can do
Ask about the problem — It will probe your use case, identify who the users are, and clarify what success looks like.
Fetch URLs — Paste a link in the chat and the AI will fetch and read the page before responding. Useful for sharing existing tools, API docs, spreadsheets, or specs.
Save structured findings — As the AI identifies requirements, open questions, data model entities, and insights, it saves them to the Findings sidebar on the right. These carry forward into the project brief when you graduate.
Analyze uploaded documents — You can attach files directly and the AI reads their full content as context for the conversation.
Uploading Supporting Documents
Click the paperclip icon in the chat input (or the + button in the Documents sidebar) to attach a file to the session.
Supported file types:
| Type | MIME / Extension |
|---|---|
.pdf | |
| Plain text | .txt |
| HTML | .html |
| Markdown | .md |
| CSV | .csv |
| Images | .jpg, .jpeg, .png, .webp, .gif |
Max file size: 10 MB per file.
Uploaded documents appear in the Documents section of the sidebar. You can remove them individually with the trash icon.
How it works internally: Document bytes are stored in R2 and re-read on every chat turn. This means the AI always has fresh access to the full file content, but it also means each message that uses documents is more expensive in tokens — keep documents focused and relevant.
The Findings Sidebar
As you chat, the AI saves structured findings to the sidebar. There are four finding types:
| Type | What it captures |
|---|---|
| Requirement | A confirmed must-have feature or behavior |
| Question | An unresolved open question that needs an answer before building |
| Data Model | An entity, relationship, or important field the app needs to store |
| Insight | An observation about the problem space, users, or constraints |
Findings accumulate across turns. You can't edit or delete individual findings from the UI — if something is wrong, just tell the AI and it will note the correction.
All findings are included verbatim in the project brief when you graduate.
Graduating to a Project
When you're ready to build, click Start Building in the header. This:
- Calls Claude Haiku to summarize the session into a concise project brief (~200 words) using the chat history and saved findings.
- Shows you the brief for review.
- On confirmation, creates a new Vulcan project with the brief stored as context — the App Builder AI reads it before your very first chat turn.
The brief is capped at the last 20 messages of the conversation. If your session is very long, the earlier context may not be fully reflected. Save key decisions as findings as you go — they're always included in full.
You can still use the graduated session for reference — it becomes read-only (no further chat turns). It stays in your sessions list, marked as Graduated.
Sharing & Collaboration
Every research session is shareable from the moment it's created. The session owner can invite teammates and manage access from the Share modal.
Inviting collaborators
- Open the session in the chat view.
- Click the Share button in the header (visible to the session owner only).
- Choose a role — Editor (can send messages) or Viewer (read-only).
- Click Generate Link — the invite URL appears in a text box and is copied to your clipboard automatically.
- Share the link with the teammate. When they open it, they're added to the session and redirected into the chat. A Copy button is also available if you need to copy it again.
Roles
| Role | Permissions |
|---|---|
| Owner | Full access — can chat, view findings, graduate, delete, manage members |
| Editor | Can chat with the AI and view all findings |
| Viewer | Read-only — can view the chat history and findings |
Shared chat thread
All session members share a single conversation thread. Messages sent by any member are visible to everyone. This mirrors how a real analyst conversation works — everyone sees the full context.
Managing members
Open the Share modal at any time to see the current member list. (Role changes and removal are coming in a future update.)
Managing Sessions
- Rename — Click the session title in the chat header to edit it inline. Press Enter or click away to save.
- Delete — Click the Delete button in the chat header, or the trash icon on the session card in the hub. Only the session owner can delete a session. Deleting permanently removes the chat history, findings, all uploaded documents, and all member access. This cannot be undone.
- Graduate — Only the session owner can graduate a session to an App Builder project.
Gotchas & Limitations
URL fetching
- 20 KB content cap — Only the first 20 KB of a fetched page is used. Long pages are truncated with a notice.
- HTML is stripped — The AI sees text content only. CSS,
<script>tags, and formatting are removed. Pages that are purely JS-rendered (SPAs) often return near-empty content. - No authentication — The AI can only fetch publicly accessible URLs. Internal Veho systems, Confluence behind SSO, JIRA, and similar require you to paste or upload the content manually instead.
- 10-second timeout — Slow-loading URLs will fail with a fetch error. Retry or paste the content directly.
Document upload
- Word docs are not supported — Export to PDF first (
File → Export as PDF). - Excel files are not supported — Export to CSV first (
File → Download → CSV). - 10 MB file limit — Larger files must be trimmed or summarized before uploading.
- Documents cost tokens on every turn — The full document is re-sent to Claude with each chat message. A 5 MB PDF costs roughly the same as a long conversation on every single turn. Upload documents when they're directly relevant and remove them when you're done with them.
- Images are analyzed for content, not parsed as data — If you upload a CSV screenshot, the AI sees an image and must OCR it mentally. Upload the actual
.csvfor structured data. - No Word, Excel, PowerPoint, or Google Docs — Only the MIME types listed above are accepted. For Office files: PDF export for documents, CSV export for spreadsheets.
Sessions & findings
- Findings can't be manually edited or deleted — If the AI saves a wrong finding, tell it in chat. You can also note corrections directly in the brief review step before creating the project.
- Session history is not infinite — The graduation brief uses the last 20 chat messages. In very long sessions, early context may be lost in the brief. Findings are always included in full regardless of session length.
- Graduated sessions are read-only — You can view the history and findings, but you can't continue the conversation.
- Owner-only operations — Only the session owner can delete, graduate, and generate share links. Collaborators can chat (Editors) or view (Viewers) but cannot manage the session.
The AI
- This is not a code generator — The Research AI is tuned for analysis and clarification. It will not write code, create files, or deploy anything. Use the App Builder for that.
- It won't access Veho internal systems — Same constraint as the App Builder. It cannot query Veho databases, Snowflake, internal APIs, etc. Paste or upload the data you need it to analyze.
- Findings are AI-generated — The AI decides what's worth saving as a finding. It may miss things or save redundant ones. Always review the findings before graduating.
- Context window limits — Very long sessions (100+ messages) may cause the AI to lose track of early context. Start a new session for a new problem rather than continuing an old one indefinitely.
Storage
- Uploaded documents count against your platform storage quota — The 100 MB per-user limit includes research document storage. Delete sessions (and their attachments) when you're done with them.
- Documents are deleted with the session — There is no way to recover a deleted session's documents.