Using Beyond mode
Beyond is the multi-step chat mode for larger compliance deliverables: plan, draft, verify, with optional grounded web research on eligible plans.
Contenido en inglés
What Beyond is
Beyond sits next to Fast and Think in the chat composer. Use it when the work is bigger than a single answer: multi-section policies, structured assessments, or research-backed drafts that should be planned before they are written.
You can also run Beyond from a connected agent (Claude Code, Cursor, Grok, OpenCode, and other MCP clients). The agent must send mode: "beyond" on create_conversation or send_message. If it omits mode, the connection defaults to Fast. Say “use Beyond” or “run this in Beyond” so the agent picks the right mode. See Connect ISMS Copilot to Claude Code.
Who can use it
- Plus and above (and active Plus trial): Beyond available.
- Free: Beyond is not available.
- Essential (grandfathered): Beyond is not part of the default Essential package; upgrade to Plus+ for Beyond.
- Advanced Data Protection: Beyond still works for chat drafting; general web discovery is restricted under ADP (exact page/document analysis can still use the EU path). See the web research article.
How to run a Beyond task
In the web app
- Open a conversation.
- Select Beyond in the mode control (next to Fast / Think).
- Describe the deliverable clearly (framework, audience, constraints, workspace files to use).
- Beyond plans and works in steps. Stay in the thread; you can steer with follow-ups.
- Export finished sections with the usual document tools when you are ready.
From a connected agent (MCP)
- Connect with a
pat-isms-…token (Connect Claude Code or any MCP client). - Ask for Beyond explicitly (for example, “use Beyond to draft an A.5.35 implementation guide”). The agent passes
mode: "beyond". - Beyond is minutes-scale. The first tool call may return
status: "generating"with abeyond_run_id. The agent should pollget_replyuntilcomplete(about every 10–20 seconds). Fast GLM-5.3 runs sometimes finish inside the first call. - Paid plans only, 10 Beyond runs per UTC day. If Beyond cannot run (plan, daily cap, or ADP), the turn answers through the normal flow and the result includes
beyond_denied_reason.
The connection does not auto-upgrade a hard question to Beyond. Fast is the default unless you or the agent set the mode.
Beyond vs Think vs Fast
| Mode | Best for |
|---|---|
| Fast | Quick Q&A, short drafts |
| Think | Harder single-shot reasoning without a multi-step plan |
| Beyond | Multi-step plan → draft → verify style work |
Provider routing (which model sits behind a mode) depends on plan, ADP, and product defaults. Do not treat any mode as “always Claude” or “always Grok.” Paid non-ADP defaults and ADP EU routing are described on the Trust Center and in Advanced Data Protection docs.
Web research inside Beyond
On eligible plans, Beyond can use grounded web research (for example company research or a specific public URL). Explicit requests such as “search the web” request the research path. Framework-stable questions still use built-in compliance knowledge without unnecessary web calls.