Running MultAI

The end-to-end workflow — how to invoke /multai, select platforms, choose a mode, and make sense of the collated output.

Overview

A MultAI run has three phases:

  1. Invoke — type /multai in Claude Code, provide your prompt, select platforms and mode
  2. Execute — the orchestrator submits your prompt to each selected platform and waits for responses
  3. Review — read the collated archive, optionally run /consolidator to synthesize findings

There are two execution paths depending on where you invoke MultAI. In the Claude Code tab all platforms run in parallel via the Python orchestrator. In the Cowork tab platforms run sequentially via Claude-in-Chrome browser tools.

A typical REGULAR-mode run takes 2–5 minutes. A DEEP-mode run with all platforms can take 15–30 minutes for the platforms that support Deep Research.

Invoking /multai

Type /multai in the Claude Code chat input. The skill automatically detects which runtime you're in (Code tab or Cowork tab) and routes accordingly.

You can pass your prompt inline or from a file:

Inline prompt
/multai --prompt "What are the key differences between RAG and fine-tuning?"
Prompt from file
/multai --prompt-file ./my-research-question.txt

If you invoke /multai without arguments, the skill will prompt you interactively for the query, mode, and platform selection.

Phase 0 detection: MultAI reads the Claude Code runtime environment to determine whether it's running in the Code tab (Playwright path) or Cowork tab (MCP path). You don't need to specify which — it detects automatically.

Platform Selection

By default MultAI runs all 7 platforms. Use --platforms to run a subset:

Run three platforms
/multai --platforms claude chatgpt gemini --prompt "Explain transformer attention"
Run a single platform
/multai --platforms deepseek --prompt "Explain transformer attention"

Platform identifiers (case-insensitive):

  • claude — Claude.ai
  • chatgpt — ChatGPT
  • gemini — Google Gemini
  • grok — Grok (xAI)
  • deepseek — DeepSeek
  • perplexity — Perplexity
  • copilot — Microsoft Copilot

Multiple identifiers are space-separated. Omit the flag entirely to run all 7.

Choosing a Mode

Pass --mode REGULAR (default) or --mode DEEP:

DEEP mode
/multai --mode DEEP --prompt "Analyze the state of quantum computing in 2026"

REGULAR mode (default)

Submits the prompt as a standard chat message. Fast — typical response time is 30 seconds to 3 minutes per platform. Runs all 7 platforms in parallel (Code tab) in under 5 minutes total.

DEEP mode

Activates extended research features on supported platforms:

  • ChatGPT — Deep Research (web search + synthesis, 5–20 min)
  • Gemini — Deep Research (Google Search + synthesis, 5–15 min)
  • DeepSeek — DeepThink extended reasoning
  • Other platforms — receive REGULAR mode prompt (no DEEP support)
DEEP mode quotas: ChatGPT Deep Research has a daily usage limit. Gemini Deep Research has a monthly cap. If your quota is exhausted, those platforms are automatically skipped. Use REGULAR mode to avoid consuming quota.

Reading the Output

MultAI writes output to two locations in your project directory:

Per-platform files — output/

One markdown file per platform, named YYYY-MM-DD-task-name-platform.md. Each file contains the full platform response wrapped in <untrusted_platform_response> XML tags.

Collated archive — archive/

After all platforms complete, all per-platform files are merged into a single archive file: YYYY-MM-DD-task-name-archive.md. This is the primary file to read — it contains all 7 responses in order.

The task name comes from the --task-name flag, or is derived from the first few words of your prompt if not specified:

Custom task name
/multai --task-name "quantum-computing-2026" --prompt "Analyze quantum computing..."

Cowork Path

When MultAI is invoked from a Claude Cowork tab, it uses a completely different execution path — instead of Playwright, it uses Claude-in-Chrome MCP browser tools to navigate each platform's UI.

Key differences in Cowork mode:

  • Platforms run sequentially — one at a time, in order
  • No Python or Playwright required on the local machine
  • Requires the Claude-in-Chrome browser extension installed and connected
  • Slower overall — sequential means total time = sum of all platform times
  • More resilient — Claude reads the screen directly, not CSS selectors
Cowork mode is the easiest way to use MultAI if you don't want to set up the Python environment. Install the Claude-in-Chrome extension, open a Cowork tab, and run /multai.

Consolidating Results

After MultAI finishes, invoke /consolidator to synthesize all platform responses into a single structured intelligence report:

/consolidator

The consolidator reads the archive file and produces a report that:

  • Identifies points of consensus across platforms
  • Highlights divergences and platform-specific insights
  • Surfaces unique information that only one or two platforms provided
  • Assigns confidence levels based on cross-platform agreement

The consolidated report is the recommended end artifact for research tasks — it distills 7 responses into one authoritative synthesis.

Previous ← Core Concepts Next Reference →