Real execution
Maximo Cloud Computer, Codex, Syntax, Hermes & OpenClaw
Maximo Cloud Computer gives Maximo a real execution environment for work that should not be guessed: terminal tasks, browser or fetch inspection, website checks, code execution, file generation, repository work, screenshots, and configured agent runs. The selected Maximo chat model remains the planner and final answerer; the built-in computer executes exact commands or scripts from that model, while connected agent profiles are autonomous runners.
- Use /computer for the built-in Maximo Computer runner. Maximo reasons first, then sends exact terminal commands or a fenced Python, JavaScript, or Bash script for the computer to execute.
- The command list is dynamic per request. MyTabulon does not hardcode a fixed command list or special-case one website, repo, language, or task format.
- If Maximo tries to send only the user's prompt to the built-in computer, the backend rejects it and asks Maximo to provide commands or a script instead.
- Use /codex after connecting a Codex profile. Codex setup uses Codex ChatGPT device login, with OpenAI API key fallback.
- Use /syntax after connecting Maximo Syntax with Maximo AI OAuth so workflows can run inside the project workspace.
- Use /hermes or /openclaw only after the matching owner/admin profile is enabled. These external agent profiles use a smaller agent-minute pool than normal computer runs.
- The slash commands are shortcuts. Maximo should also call the computer tool naturally when a user asks in any language to scrape, run, test, inspect, generate files, use the computer, or use a named agent.
- Project folders keep uploads, generated outputs, code, screenshots, logs, run files, Agent Runs, and Cloud Workspace snapshots together. When Cloud Workspace is enabled on the project, later Maximo Computer, Codex, Syntax, Hermes, or OpenClaw runs restore the same logical workspace before starting and save compatible changes back after the run.
- The absolute runner path can change between cloud computers, so Maximo should refer to the attached project Cloud Workspace or repo path instead of presenting a temporary runtime path as the durable location.
- The Files page can open Workspace Files for the saved cloud workspace tree. Users can browse cloned repositories, move through folders, download files, show hidden system workspace folders when needed, and edit text/code files in a compact IDE-style editor.
- The Workspace Files editor supports syntax highlighting, line numbers, copy/download/save, wrap on by default for long lines, and expand/collapse for full-screen editing on desktop and mobile.
- Project cards and project edit/create controls show Cloud Workspace status and the enable/disable toggle. Hidden workspace snapshots stay out of normal file browsing until the hidden-files toggle is enabled.
- Workspace snapshots and visible project files count against the plan storage quota. Storage warnings should be soft at high usage, with cleanup suggestions for temporary logs, caches, screenshots, ZIPs, and build artifacts.
- Computer runtime is metered separately from model tokens. The chat model's planning and final answer count against normal AI usage, while browser/code/agent runtime counts against computer or external-agent minutes.
- If an agent profile uses MyTabulon-managed model providers, those model calls should count against the user's AI usage. If the user connects their own external API key directly, provider billing can happen on their key while MyTabulon still meters runtime.
- Settings for Codex, Maximo Syntax, Hermes, and OpenClaw live under Integrations and are owner/admin-only because credentials, network access, Git write access, deployment, and external tools can be sensitive.

