How to Use Metallm AI Aggregator

Sign Up and Open the Chat

Go to metallm.tech/chat and create a free account. Once signed in, you will see the main chat interface with a sidebar on the left showing your chat history, and a message input at the bottom.

At the bottom of the screen, you will notice the mode selector. By default it is set to Direct mode. Next to it is the model selector showing which AI model you are currently chatting with (for example, GPT-5.4).

The available modes are:

Select a Model and Start Chatting

In Direct Mode, click the model name at the bottom (for example "GPT-5.4") to open the model selector. You will see a list of all available models organized by tier:

• Claude Sonnet 4.6 — Advanced Reasoning and Technical Writing
• Gemini 3.1 Pro Preview — Long Context Reasoning, Research and Multimodal
• Grok 4.20 309 Reasoning — Social Signals and Reasoning
• Claude Opus 4.7 — Top Tier Agents, Coding and Deep Analysis
• GPT-5.5 — Reasoning and Professional Workflows
• Grok 4.3 — Real-Time Search, Social Signals and Reasoning

Pick any model and start typing. The selected model will respond to your message. You can switch models at any time during the conversation, and the new model will be aware of the entire previous conversation — no context is lost.

Direct Mode Response

When the model responds, you see the model name and badge at the top of the response (for example "GPT-5.4 | Advance Reasoning and Professional Workflows"). The response appears directly in the chat with full markdown formatting.

In this example, GPT-5.4 explains what Metallm AI Aggregator is and lists all its core features — Direct mode, Smart Route, Multi mode, Debate mode, shared conversation memory, and more. The response is clean, detailed, and contextually aware.

Let AI Pick the Best Model and Enhance Your Prompt

Switch to Smart Route from the mode selector at the bottom. Now when you type a message and send it, something different happens:

1. The system analyzes your prompt to determine which AI model is best suited for this task
2. It routes your question to a specialist model (you see "Routing to Specialist" at the top)
3. Your prompt gets enhanced — the system rewrites it to be more detailed and effective
4. A preview appears showing the enhanced prompt, the selected model, and your original prompt

In the preview, you can see:

• The Enhanced Prompt (improved version of what you typed)
• Your original prompt for reference
• The Selected model that was chosen (for example, Qwen 3.6 Plus for a Python coding task)
• A dropdown to change the model if you disagree with the selection

You have three options: Cancel, Send Original (use your original prompt as-is), or Approve and Send (use the enhanced prompt with the selected model).

Configure Your Debaters

Switch to Debate mode. Click "Debate Config" at the bottom right to open the configuration panel. Here you set up how the debate works:

• Debate Rounds — choose between 3 to 5 rounds (minimum 3). After the final round, a neutral judge model provides a verdict
• Debater 1 — select a model (for example GPT-5.4) and assign a role/perspective (for example "Angel")
• Debater 2 — select a different model (for example Claude Sonnet 4.6) and assign a different role (for example "Devil")
• System Prompt (optional) — add custom instructions for each debater to guide their perspective

The page heading says "Let Them Debate" with the description: "Models argue your question from different angles — get richer, more nuanced insights." You will see both debater role badges at the bottom of the chat input.

Preview Before Sending

When you type your question and hit send, a preview panel appears showing:

• The heading "Debate Mode" with "Per-Model Prompts (edit before sending)"
• GPT-5.4 with the "Angel" badge — showing the prompt it will receive
• Claude Sonnet 4.6 with the "Devil" badge — showing the prompt it will receive
• Your original prompt for reference
• Confirmation of which models will debate

You can edit each model's prompt individually before sending. This means you can tweak how each model sees the question to get more interesting debates. Then click "Approve and Send" or "Send Original" or "Cancel".

Watch the Debate Unfold

Once approved, the debate begins. Each model takes turns responding according to their assigned role. In this example:

The Angel (GPT-5.4) argues the positive perspective in its style — direct, punchy, motivational. It talks about how "good work" builds discipline and creates real power.

The Devil (Claude Sonnet 4.6) fires back with "OPENING ARGUMENT" — arguing that "Good work is a cage dressed as a crown" and that every empire was built on breaking rules, not following them.

At the bottom you see "Round 1 of 3 complete" with options to browse rounds, give instructions for the next round, or request a verdict early. After all rounds complete, a neutral judge model provides a final verdict summarizing both sides.

Select Multiple Models

Switch to Multi mode. The heading says "All Models, One Query" — every model answers simultaneously so you can compare perspectives side by side.

Click "Select Models" to open the model picker. You will see all available models organized by tier (Premium and Balanced) with checkboxes. Pick as many models as you want — each selected model will answer your question independently.

Models available include:

• Premium: GPT-5.4, Claude Sonnet 4.6, Gemini 3.1 Pro Preview, Grok 4.20 309 Reasoning, Claude Opus 4.7, GPT-5.5, Grok 4.3
• Balanced: Claude Haiku 4.5, GPT-5.4 mini, Memo V2 Pro, Gemini 2.5 flash

At the bottom of the picker you see a count (for example "6 models") and a "Select All" option if you want every model to respond.

Per-Model Prompt Enhancement

After you type your question and send it, Multi Mode shows a preview with per-model prompts. The system enhances your prompt differently for each model based on that model's strengths:

• Claude Haiku 4.5 (Fast Summaries, Support and Lightweight Coding) — gets a prompt focused on fast, clear summary
• GPT-5.4 mini (Cost-Optimized Coding, Agents and Daily Assistant) — gets a practical, execution-focused prompt
• Grok 4.3 (Real-Time Search, Social Signals and Reasoning) — gets a prompt emphasizing current market signals and competition
• Qwen 3.6 Plus (Strong Reasoning and Multilingual) — gets a prompt asking for careful reasoning and balanced conclusion

You see your original prompt at the bottom, and confirmation of which models will receive it. You can edit any individual model's prompt before sending. Then click "Approve and Send" to send to all selected models at once.

Unified Summary of All Responses

After all models respond, Metallm generates a Unified Answer — a summary that combines the best insights from every model's response into one cohesive answer. You will see tabs for "Summary" and "Final Summary" at the top.

The unified answer is smart and synthesized. In this example, the user asked "Should I start an AI startup today?" and the summary combines all 4 models' perspectives into actionable advice. It references specific points from different models and gives a clear recommendation.

You can also expand each individual model's full response to read it separately. The summary makes it easy to get the key takeaway without reading 4 separate long answers.

Select an Image Model

Switch to Image mode using the mode selector. Click the model name to see the available image generation models:

• GPT 5.4 Image 2 — Top AI Generation with Text Understanding (selected by default)
• Gemini 3 Pro Image — Advanced Image Generation with Text Rendering
• Flux 2 Pro — High-Quality Image Generation with Strong Prompt Following
• Grok Aurora — Fast and Affordable Image Generation by xAI
• Grok Aurora Pro — Premium High-Quality Image Generation by xAI

Below the model list, you can choose the Aspect Ratio for your image: 1:1, 2:3, 3:2, 2:4, 4:3, 9:16, 16:9. You also see resolution options (1K, 2K, 4K).

The input field at the bottom says "Describe the image you want to generate..." — type your description and send it.

Generate Your Image

Type a description of the image you want (for example "create a image of cat in space") and hit send. You will see a loading state with "Generating... Creating your image..." while the AI creates your image.

The generated image appears directly in the chat. The model name and a badge are shown above the image (for example "grok-imagine-image" with an "image" badge). You can generate as many images as you want in the same conversation.

At the bottom you can switch between Image and Video tabs, and change the model for the next generation.