Radar, November 13, 2026: ten signals from X on Claude Meetings, fenced-in agents and Perplexity Decider

By Ivo Donker · Friday November 13, 2026 — compiled by Claude (AI, Anthropic) from X search briefings by Grok Bot (AI, xAI) and the llmnet team's X trend monitor

All daily signals · Markdown version

In short: The standout signal of November 13: Claude can join Google Meet as a visible participant, in beta for Team and Enterprise, with a transcript and no audio. Alongside it, ten short signals on agents on your own infrastructure, how OpenAI fences in its OpenClaw agents, open decision models from Perplexity and Anthropic's IPO, each with its source on X.

Illustration: a radar screen on a desk with glowing blips: a video call, a fenced-in container, a small chip and a stock chart.

Illustration: AI-generated (OpenAI image model via Codex): a radar screen with ten signals.

1.Claude Meetings: Claude joins Google Meet, in beta for Team and Enterprise

Confidence: high (help article read)

Anthropic lets Claude join Google Meet calls as a visible participant. Claude takes notes and action items, answers in the chat and builds documents or whiteboards. A transcript is kept, no audio. On Enterprise Claude stays out of calls with external guests by default; on Team it may join by default.

Source on X: @claudeai · Claude Help Center

Context in the briefing: Claude Meetings in the November 13 briefing

2.Claude Managed Agents on your own infrastructure: private beta

Confidence: medium (release notes via search snippet)

Anthropic offers Managed Agents inside an organization's own security perimeter: agents, sessions and history stay with the customer, model requests go to the chosen provider, Anthropic by default. It is a private beta through the account team. We only saw the release notes themselves as a search snippet.

Source on X: @summer_ikiki_ · Claude, product-updates

3.OpenAI describes how it fences in hundreds of OpenClaw agents

Confidence: high (blog read)

On its developer blog (November 12) OpenAI describes its internal setup: gateway and agent in separate containers, a sandbox for network, write access and processes, secrets through a credential broker, a second agent reviewing every action, a service account per agent and logs of every turn. The danger-full-access mode is disabled.

Source on X: @kevins8 · OpenAI Developers

Context in the briefing: Agent management in the November 13 briefing

4.OpenClaw cut answers off at 8,192 tokens without maxTokens

Confidence: high (PR read)

Without maxTokens in the model config, OpenClaw filled in 8,192 output tokens for unknown models; long agent turns stopped in the middle of a tool call. Pull request #193899 (November 6) raises that to up to 32,768. Saved old values still apply: set a limit per model yourself.

Source on X: @joelniklaus · GitHub, PR #193899

Context in the briefing: Agent management in the November 13 briefing

5.Perplexity Decider v1.3: open decision models with locate

Confidence: high (model card read); free end date from X only

Perplexity released pplx-decider-v1.3 (based on Qwen3.8-27B) and a 4B Flash version under Apache 2.0. They return probabilities instead of text; new is locate, which extracts verbatim spans from the input. Useful as a cheap router or filter in front of a large model. Free until November 19 on OpenRouter according to Perplexity on X.

Source on X: @perplexitydevs · Hugging Face

Context in the briefing: Perplexity Decider in the November 13 briefing

6.Gemini 4 Hydrogen in Cursor and OpenClaw on day one

Confidence: medium (posts via the radar briefing; price confirmed by Google)

A day after launch, Gemini 4 Hydrogen can be selected in OpenClaw and in Cursor, at the same 1.50 and 7.50 dollars per million tokens as at Google. Cursor cites 54.3 percent on its own CursorBench against 39.6 for Gemini 3.8 Flash; we did not check that figure on a Cursor page.

Source on X: @openclaw · @cursor_ai

Context in the briefing: Gemini 4 Hydrogen in the November 13 briefing

Read also: Gemini 4 Hydrogen · Models and prices

7.Anthropic roadshow: 70 to 78 dollars per share, ticker ANT

Confidence: high (release read); trading date unconfirmed

According to the November 10 press release, Anthropic is offering 825 million Class A shares at 70 to 78 dollars, with an application to list on Nasdaq as ANT. The registration is not yet effective. A first trading day of November 19 appears only in X posts and secondary reports, not in the release.

Source on X: @AnthropicAI · Anthropic

Context in the briefing: The IPO in the November 13 briefing

8.Hermes Agent: catalog grows, no new release

Confidence: medium (release status via the radar briefing; plugin count unconfirmed)

Community apps keep appearing around Hermes Agent, such as Herald for iPhone and Apple Watch, and the plugin market is said to have passed 2,000 plugins. The latest release is still v0.23.1 of November 10. A bigger catalog means more third-party code: review a plugin before you install it.

Source on X: @HermesAgentTips · @iamlukethedev · GitHub, releases

9.pxpipe: system prompt as an image to save tokens, unverified

Confidence: low (X only)

A post describes pxpipe, an open-source proxy that sends Claude Code's fixed system prompt as a PNG; a session supposedly drops from 42.21 to 6.06 dollars. We did not check the repo or the figures. A proxy between your harness and the API sees all traffic, including code and keys.

Source on X: @bebuu_ai

10.No GPT-6.2 yet, and a wave of near-identical posts about Simular Box

Confidence: low (absence and pattern on X)

After the OpenAI livestream on cyber no new model appeared; GPT-6.2 is still a rumor. At the same time almost the same text about Simular Box showed up at several accounts, and Peter Steinberger warns that such posts are being paid for. Treat both as unconfirmed.

Source on X: @kimmonismus · @steipete

Context in the briefing: Unconfirmed in the November 13 briefing