Grok Bot worker-bot post
Lead: The post is a 4-minute animated tutorial for the Grok Bot app. A dispatcher bot hands coding and image work to a dedicated "worker" bot. The worker drives Codex CLI, or OpenCode with free models, on a 24h cloud computer or on your own Mac. Allternit already has most parts. The gap is the chat-driven setup: the bot installs and signs in the CLI by itself.
What the post says
- Author
- @ChuckBass14291 (Chinese AI-agent account, 7.9k followers)
- Posted
- 2026-10-28
- Reach
- 74.8k views, 828 likes, 1,098 bookmarks
- Media
- 4:09 video, animated mock-up (not real screen footage; the login URL is auth.example.com)
- Claim
- one bot can direct Codex, Claude and Gemini without using its own chat quota
Translated text, in three points:
- Main/worker split. Make a dedicated "Codex worker" bot. The main agent dispatches; the worker does the job. Tasks and context stay separate.
- 24h cloud vs local. Authorize Codex on the cloud computer with a ChatGPT device code. Your own computer can sleep while the cloud one writes code, builds pages and calls GPT Image. If you care about data, use local authorization: a pop-up asks before each run.
- Free compute. Connect OpenCode with free models (for example DeepSeek V4.1 Flash Free). The post warns that the free model list changes.
The flow in the video
The last shot shows a "tool-worker team" group in the bot list. It holds an opencode worker and a Codex worker. They pass results to other bots, such as a video monitor and a cover maker.
Mapped to Allternit
| Grok Bot feature | Allternit today | Note || 24h cloud computer | ok Cloud computer per paid user (live since 2026-10-01) | Same idea || Local computer option | ok Local runtime stays (decision 13) | Same idea || CLI as the brain (Codex, Claude) | ok CLI brains + subagent catalog (#1915) | Same idea || Bots hand work to other bots | ok Agent network / Factory | Same idea || Bot installs + signs in the CLI from chat | warn Not checked | Plan says "Connect a model" does this from Settings, not from chat || Device code relayed in chat | warn Not checked | Needs safety rules (panel D) || Free OpenCode models as a worker | warn Not checked | Cheap always-on worker || "Ask before running" pop-up for local | warn Not checked | Approval per run |I did not open the Allternit code for the four "Not checked" rows. To check: search allternit-ai and allternit-platform for device-code login and an OpenCode adapter.
Risks to keep in mind
The video's own OpenAI page warns that device codes are a common phishing target. A bot that posts "open this link, type this code" teaches users to trust that message. If Allternit copies this, show the code only from verified CLI output, allow only the vendor's own domain, and mark the message as system-generated.
- Signing in a vendor CLI on a cloud computer is the user's own login on a machine they rent. It is not the vendor-token reuse that was dropped on 2026-10-16 (SIWC). Still, check each vendor's terms for hosted use before marketing it.
- "Free models" is not a promise you can sell. The post says the list changes; keep it as an option, not a plan feature.
- "No chat quota used" is true only because the CLI uses the user's separate subscription quota.
What is worth taking
- Worker-bot setup in one chat message. "Make a bot that owns X; other bots send X to it." This is the strongest idea for non-technical users.
- "Connect Codex" from chat, using the same manifests as Settings → Subscriptions.
- A free OpenCode worker as the default brain on free cloud computers.
- The video format. A clean, calm explainer with a mascot and subtitles got 1.1k bookmarks. It is a good reference for Gizzi feature films.