How AI routing works in ISMS Copilot
Modes (Fast, Think, Beyond), plan access, Advanced Data Protection, and automatic provider routing. There is no multi-vendor model picker in chat.
ISMS Copilot does not expose a sidebar list of third-party models (Claude / GPT / Gemini / Grok) for you to flip per message. Routing is automatic based on your plan, mode, and Advanced Data Protection (ADP). You choose how hard the product should work (Fast / Think / Beyond) and whether ADP is on; the product selects the live provider path.
Older marketing or Ferndesk pages that said "open Assistants and click Claude or Grok" are stale. Treat this page and the Trust Center as current.
What you control in the product
| Control | What it does |
|---|---|
| Fast / Think / Beyond | Depth and shape of the run (modes guide) |
| Web research | Plus+ (not Free / default Essential); may be restricted under ADP discovery rules |
| Advanced Data Protection | Prefer EU-residency AI path (ADP) |
| Plan / trial | Unlocks Think, Beyond, credits, and upload fair use (pricing) |
Plan access (modes)
- Free: Fast only
- Essential (grandfathered): Fast + Think; Beyond / web research not in the default Essential package
- Plus and above (or active Plus trial): Fast, Think, Beyond, web research (subject to ADP and usage windows)
Provider routing (mid-2026 product truth)
Examples of live routing (verify details on the Trust Center if you need contractual language):
| Situation | Typical path |
|---|---|
| Paid Plus+, ADP off | xAI Grok zero-retention style path as default for Fast / Think / Beyond, with Anthropic backup if needed |
| ADP on | EU Mistral topology for chat AI processing |
| Free / Essential | Economy routing (OpenRouter-class paths); not the same default as paid Grok |
Do not write customer-facing claims that "every message always uses Claude" or that "Think is always Opus." Both are false under current defaults.
How to pick the right mode for a task
| Task type | Prefer |
|---|---|
| Quick Q&A, short drafts | Fast |
| Hard single-shot analysis, long reasoning | Think |
| Multi-step plan → draft → verify | Beyond |
| Fresh external facts / URLs | Web research (Plus+) or Beyond research where enabled |
| Strongest data-residency posture | Turn ADP on, accept Mistral path tradeoffs |
Compliance knowledge injection and framework skills still apply across routes. Mode and ADP change depth, latency, and residency posture, not "whether the product knows ISO 27001."
Over-quota and fallback
On some paid plans, when you hit the 4-hour usage limit the product may offer to continue on a faster overflow path until the next fixed UTC bin (ADP stays on Mistral; Essential is not the same path). See in-app messaging and What to do when you hit your usage limit.
Competitor-style comparisons
Narrative pages that compare "using Claude vs GPT for compliance" are about general model strengths, not an in-app model menu. For product how-to, stay on modes + ADP + plans.