Search & Research Modes
In the chat input bar SKH offers up to five research modes. Which ones are visible depends on your organisation's configuration.
Modes at a glance
| Mode | Description | When to use |
|---|---|---|
| Internal | Search only your own documents | Default for knowledge work, confidential content |
| Web | Live web search with source attribution | Current topics, public knowledge |
| Hybrid | Combines internal + web in one answer | Research mixing internal & external context |
| MCP | Calls into your tools (custom integrations) | Structured data, your own APIs (Jira, GitHub, …) |
| Deep Research | Agent breaks the question down and researches in multiple steps | Complex research, market scans, multiple sources |
Internal – Search your own documents
The default in most setups. SKH:
- Understands your question.
- Searches your knowledge base for the most relevant passages (across every workspace you're a member of).
- Hands those passages to the model as context.
- Generates the answer and produces citations to the underlying files.
Per answer you see:
- Citation cards – Click opens the source modal or the PDF viewer.
- "Show all sources" – List of every passage the model used.
Focusing a workspace
By default SKH searches across every workspace you can see. To focus a specific page or workspace, mention it explicitly in your question ("In the workspace 'Contracts': what does clause 4.3 say?").
Web – Live internet search
When enabled, SKH uses a web search and integrates the top results into the model context.
- Sources – Each web hit appears as a clickable source (URL, title, snippet).
- Cache – Results are cached organisation-wide so the daily quota stretches further.
- Daily limit – Per organisation a daily limit applies (default: 50 searches/day). Resets at midnight.
The limit is organisation-wide. If one member runs 50 searches in a day, everyone else sees a "limit reached" notice for the rest of the day.
Hybrid – Internal + web
Best of both worlds: SKH first builds context from internal passages, then augments with web results. Answers cite both source types with distinct badges.
When to use?
- "Compare our internal contract terms with current industry standards."
- "Where do our docs differ from OpenAI's docs on function-calling?"
MCP – Tool-based chat
Model Context Protocol (MCP) is an open standard for "persistent tools" the AI model can invoke at runtime. Examples:
- Jira server – Read/write issues.
- GitHub server – Pull requests, issues, code search.
- Your own API server – Query ERP data.
When you enable MCP mode, the model sees the tools you've been granted and decides on its own when to call which. Tool calls appear in chat as collapsible cards.
Prerequisites:
- The server is registered and Active.
- You (or your group) have permission for the tool.
Details: MCP Servers.
Deep Research – Multi-step research
Deep Research is its own research agent that decomposes an open question into multiple sub-steps:
- Plan – The agent analyses the question and lays out a research roadmap.
- Iterate – For each sub-step it runs web searches, internal queries and possibly tool calls.
- Synthesise – Finally it consolidates the findings into a structured answer.
In chat you'll see a live progress display: which searches are running, what sources have been found, which sub-step is active.
Deep Research can take several minutes and consume more tokens than a standard chat. Use it for:
- Market analyses.
- Technology evaluations.
- Multi-part research questions with unclear data.
Organisation LLM scope
Owners set under Settings → LLM Scope which modes are available:
- Internal only – Default. Only internal search.
- Full access – You additionally pick: Web, MCP, Hybrid, Deep Search.
Practical recommendations
- Government, healthcare, legal – Internal only or Internal + MCP.
- Marketing, sales, R&D – Full access with Web + Hybrid.
- Consulting, analysis – Full access with all options.
Global search vs chat search
Don't confuse them:
- Chat search (modes above) – Answers questions with a model. Surfaces sources.
- Global search (Ctrl/Cmd + K) – Classic keyword search through files, pages, folders, workspaces, chats, messages. Returns direct hits without an AI answer.
They complement each other: global search gets you to a specific document quickly, chat search produces an explanatory answer from it.
Tips for great research
- Be specific – "What are FADP requirements for data processors?" outperforms "What is FADP?".
- In web mode, mention the date – "As of 2026, what are the EU's planned GDPR updates?" helps the model prioritise recent hits.
- In hybrid, compare sources – Look at badge colours per citation to distinguish internal vs external.
- Frame Deep Research concretely – "Build a competitive comparison of the top three CRM vendors in the SMB market in 2025, with strengths, weaknesses and pricing" works better than "what's the best CRM?".