LLM Configuration
This page targets owners and describes the three AI-related areas under Settings → AI & Data:
| Area | Path | Who edits | What is set |
|---|---|---|---|
| LLM Scope | Settings → LLM Scope | Owner / Admin | Which research modes (Internal, Web, MCP, Hybrid, Deep Research) members may use |
| Model Policy | Settings → Model Policy | Owner / Admin | Which BYOK providers are allowed in the organisation, plus an optional default model |
| Model Profiles (BYOK) | Settings → Model Profiles | Each member for themselves | Personal profiles with your own API key, provided the model policy allows that provider |
A fourth entry in the same group – MCP Servers – has its own guide: MCP Servers.
LLM Scope
Controls which search and research modes are available to your members in chat.
Levels
| Level | Meaning |
|---|---|
| Internal only | Internal document search only. Default for new organisations. |
| Full access | You additionally pick: Web, MCP, Hybrid, Deep Search. |
Examples:
- Internal only – Ideal for government, legal, healthcare.
- Full access with Web + Hybrid – Sales, marketing, R&D.
- Full access with all options including Deep Search – Consulting, research.
Configure
- Settings → LLM Scope.
- Pick a level.
- With Full access, pick the additional options.
- Save.
Members see the chat search-mode toggles only after a reload.
Model Policy
Defines which BYOK providers are allowed in your organisation, and optionally an organisation-wide default model.
Configuration
- Settings → Model Policy → Policy tab.
- SKH Standard is always on. The included models (DeepSeek V3, Kimi K2.5, Mistral Medium 2505) ship with every plan.
- BYOK providers – Toggle per provider:
- OpenAI, Anthropic, Google Gemini, Mistral AI, Azure OpenAI, Azure DeepSeek, Custom OpenAI-compatible.
- Optionally set a standard default model. New chats then start with that model.
- Save.
Tip: keep BYOK off initially. Only enable a provider when there's a concrete need – every enabled provider is a potential outbound data path.
Audit
The Audit tab shows every switch from an SKH Standard model to a BYOK model in the last 30 days:
- Who (member name + e-mail) switched?
- When?
- To which provider and profile?
- In which chat?
Logs cannot be deleted.
Personal Model Profiles (BYOK)
If your organisation has BYOK enabled in the model policy, every member can create their own personal model profiles.
Creating a profile
- Settings → Model Profiles.
- Add profile.
- Pick a provider (only allowed providers are listed).
- Enter a model ID (e.g.
gpt-4o,claude-3-opus-20240229,gemini-1.5-pro). - Paste your API key. The key is stored encrypted.
- For Azure providers also: endpoint, API version, deployment name.
- For Custom OpenAI-compatible: enter a base URL.
- Optionally mark the profile as your personal default.
Profiles are private – other members can't see your API keys or use your profiles.
Using in chat
Your BYOK profiles appear as additional options in the chat's model selector, alongside the SKH Standard models. The first time you switch to a BYOK profile, SKH shows a warning; an amber strip stays pinned at the top of the chat as long as the profile is active.
Details: AI Models & Data Protection.
Watch your usage
The Dashboard and Billing → Usage show aggregated metrics that help you assess the effects of your configuration:
- Token usage per member and model, including estimated cost.
- Web searches per day with cache hit rate.
- Knowledge-base storage consumed (in GB).
Best practices
- Start restrictive – Internal only + no BYOK. Expand only when members raise a concrete need.
- Roll BYOK out gradually – Enable a single provider first, evaluate the data flow via the Audit tab in Model Policy, then expand.
- Audit regularly – Frequent switches to external providers should be justifiable.
- Track cost – With BYOK you bill directly with the provider. Token usage on the dashboard helps sanity-check the external cost.