Run Beyond from the editor you already work in

A lot of GRC work now happens in the same editor as the rest of the job. You keep Claude Code, Cursor, or Codex open. You ask questions there. Until recently, ISMS Copilot could meet you there for a Fast or Think answer, through the account connection described in Connect ISMS Copilot to Claude Code. The mode built for the bigger job could not.

That bigger job is a set, not a question. A missing GDPR document pack. Audit notes that have to become a findings report. Beyond exists for that: it plans the work, drafts each document as its own step, and verifies each one against the plan before you see it. The bottleneck on that work is not producing text. It is trusting it.

So the failure mode was simple. The work that most needed checking was the work that forced you out of the editor. You ran Fast from the agent, got a plausible draft, and either accepted it unchecked or opened the web app, switched to Beyond, waited, and brought the result back. Two copies. Two contexts. Moving drafts between tools can create competing copies. Running Beyond from your connected agent reduces that manual handoff.

ISO/IEC 27001:2022 (published October 2022) clause 7.5.3 requires documented information to be available where it is needed, adequately protected (including from improper use), and under control of changes. The standard does not require that checking happen inside any particular tool.

What we shipped

Connected agents can now run Beyond on your conversations.

If you already connect ISMS Copilot from Claude Code, Cursor, Codex, or another MCP client, the agent can send a turn in Beyond: it plans, drafts, and verifies, then returns the deliverable on that conversation. Eligibility matches Beyond in the web app. Ask your agent to use Think or Beyond. It must pass mode: "think" or mode: "beyond"; the server defaults to Fast when mode is omitted. The server does not auto-upgrade a hard question.

Longer runs often come back as still working. When the tool returns generating, the agent should poll get_reply for the result. If a run stops early, the result identifies that status and can include partial deliverables. If Beyond cannot run on that turn (eligibility, a daily cap, or another gate), the turn answers through the normal flow rather than failing silently, and the result says so.

This shipped in September 2026. It is live. Setup, scopes, and the connect command live in the agents docs and in Using Beyond mode. This post is not that guide.

Before and after

Before, the account connection could hold a real ISMS Copilot conversation from the editor, in Fast or Think. Beyond sat next to the message box in the web app. A multi-document job meant leaving the thread you were in.

After, the same connection can run the plan-draft-verify mode. You stay in the editor. The conversation and the deliverable stay on your account, on the same chat subscription as the web app. You do not paste a finished pack from one window into another.

What this is not

Three easy confusions, because the product surface has more than one door.

• Not the model API. Agents use a personal access token and act as you inside your account (conversations, workspaces, memories). The model API is a different credential and a different job: completions for code you write. Different doors, different keys. The split is on /products/agents.
• Not Agent Tasks. There is a separate in-app Tasks surface. It is not generally available, and it is not this. This post is about Beyond on a connected agent, the same plan-draft-verify mode as the web composer.
• Not an automatic upgrade. Fast is the default if the agent omits the mode. The agent has to pass mode: "think" or mode: "beyond". The server will not decide you meant the slower path.

ChatGPT connectors that only speak OAuth cannot use a personal access token yet. Claude Code, Cursor, Codex, and other HTTP MCP clients can.

Who it is for

• People whose GRC work already lives in an editor, who were leaving it to run Beyond in the browser.
• GRC engineers who keep the coding agent for orchestration and want the specialist for the compliance step, the job described on /for/grc-engineers.
• Anyone who already connected their account and has been using Fast from the agent, waiting for the checking mode to follow.

If you only ask one-off questions, Fast is still the right default. Beyond earns the extra steps when a single pass would leave you doing the verification yourself. That was true in the web app. It is true from the agent. The change is that you no longer have to pick a different window to get it.

Try it

If you are already connected, ask the agent to use Beyond on the next job that is a set of documents rather than a single answer. If you are not connected yet, create a token under Settings, Connected apps, and follow the connect guide. Then run the job in the editor you were going to use anyway.

The checking was always the work. It can happen where the rest of the work already happens.