Skip to main content

​
Research Studio

Research Studio is a three-column workbench that lets you curate research materials, converse with AI, and preview generated artifacts—all in one interface. It works on local WebUI, Tauri desktop, and SaaS cloud.

​
Layout

On mobile devices (viewport ≤ 768 px), the three columns collapse into tab navigation for comfortable use on phones and tablets.

​
Open Research Studio

Navigate to /research from the sidebar or browser URL bar. The page loads with the Resource Pool on the left, Chat in the center, and Output on the right.

​
Add research materials

​
From Wiki

  1. Click From Wiki in the resource pool header.
  2. Search for knowledge base concepts by keyword.
  3. Click a concept to add it to the pool. Duplicates are prevented automatically.

​
Upload files

  1. Click Upload in the resource pool header.
  2. Select one or more files (PDF, DOCX, TXT, MD, CSV, XLSX, PPTX) — up to 20 files at once, 50 MB per file.
  3. Uploaded files appear in the resource pool as selectable items.
  4. Large documents (> 100 KB) are automatically indexed into the knowledge base for intelligent RAG retrieval, saving LLM tokens.
  5. Duplicate files are detected via SHA-256 hashing and skipped automatically.

​
Select and deselect

Each resource has a checkbox. Selected resources are automatically injected into the AI chat context as @wiki_concept or @wiki_raw_file mention references.
  • Select All / Deselect All — toggle all resources at once.
  • Individual toggle — click a resource row to toggle its selection.
Only wiki-type references are managed by the sync. Any other mention references you add manually in the chat input (like @file.pdf) are preserved.

​
Chat with context

The center column is a full chat interface. When you select resources in the pool, their content is injected into the next message you send. The agent can reference and reason over all selected materials. Use the built-in builtin-researcher agent or any custom agent for specialized research tasks.

​
View and export artifacts

When the agent generates artifacts (reports, slide decks, documents, mind maps, audio), they appear in the right column:
  • Preview — rendered content with syntax highlighting, interactive diagrams, and media playback.
  • Download — save the artifact to your local device.
  • Save to Wiki — persist the artifact directly into the knowledge base for future reference.
  • Tab switching — when multiple artifacts are generated, switch between them using the tab bar.

​
Supported output types

​
Research → Notes → Presentation workflow

Research Studio supports the complete research-to-deliverable pipeline without leaving the workbench:
  1. Research — Select materials in the resource pool and ask the agent to conduct deep research. The builtin-researcher agent supports HITL plan confirmation, where you can review and edit the research plan before execution.
  2. Save as note — When the agent produces a report, click the Save to Wiki button on the message action bar. Notes are saved with provenance metadata (source chat, message ID, timestamp), so you can always trace back to the original conversation.
  3. Edit — Open the saved note in the Wiki knowledge base panel and edit the Markdown content directly.
  4. Generate presentation — Ask the agent “Create a PPT from this report” in the chat. The office-document skill generates fully editable .pptx files using python-pptx in a sandboxed environment. The agent has full control over slide content, layout, charts, and styling—no fixed template constraints.
  5. Download — The generated presentation appears in the right output column, ready for download.
This conversational workflow is more flexible than fixed pipeline engines: you can intervene at any step, ask the agent to revise specific sections, change the presentation style, or restart from any point—all through natural conversation.

​
Cross-document analysis

Upload multiple files to the resource pool (or import them into Wiki), select them all, and ask analytical questions like:
  • “What consensus views do these reports share?”
  • “Where do these documents contradict each other?”
  • “Which report originally stated that AI will replace 30% of jobs?”
The agent reads all selected documents, identifies consensus, contradictions, and attributed quotes, and produces a structured analysis report. Key features:
  • Clickable citations — Every attributed quote links back to the original source text via Citation Source Jump. Click a citation to see the highlighted passage in context.
  • Parallel analysis — For large document sets, the agent can delegate analysis of individual documents to sub-agents in parallel, then synthesize the results.
  • Information integrity — Built-in integrity rules ensure claims are evidence-based and contradictions between sources are explicitly flagged.
  • Flexible angles — No preset analysis templates. Ask any question—trends, timelines, sentiment, risk factors—and the agent adapts its analysis dynamically.

​
Responsive design

The workbench adapts to your screen: Column widths are adjustable via drag handles between panels.