3.3 Multi‑Model Management (Model Manager)

Pop provides a powerful and flexible multi‑model management system, allowing you to freely switch between various model providers such as OpenAI, DeepSeek, Ollama local models, and custom enterprise API models. This chapter explains how to configure, manage, switch, and optimize these models.

🎯 Why Do You Need Multiple Models?

Different AI models excel in different task scenarios:

Model	Best At
OpenAI (GPT series)	General ability, reasoning, stability
DeepSeek series	Math logic, coding, stronger reasoning
Local models (Ollama/LM Studio)	Offline use, privacy, low cost
Enterprise self‑hosted models	Data privacy, custom fine‑tuning
Multi‑vendor API pools	Choose the optimal model per task

Therefore, Pop allows you to configure multiple models simultaneously and switch freely in each session/window.

🧱 Model Management Interface Structure

The model manager typically includes:

Model List (configured models)
Add New Model
Model Details / Edit
Global Default Model
Enable / Disable Switch

You can view all model statuses, test connectivity, and adjust settings.

🛠 Model Types & Configuration

Pop supports the following categories, each with slightly different configuration requirements.

1) OpenAI Models

Supported:

GPT‑4 series
GPT‑4o / GPT‑4o mini
GPT‑3.5 series
Custom OpenAI‑compatible endpoints

Required Fields:

API Key
Base URL (optional, for OpenAI‑compatible services)
Model name (e.g., gpt-4o)
Default parameters (temperature, max_tokens, etc.)

Optional Features:

Proxy support
Rate‑limit protection

2) DeepSeek Models

Supported:

DeepSeek‑R1
DeepSeek‑V3
All DeepSeek official API models

Required Fields:

API Key
Base URL (optional)
Model name (e.g., deepseek-r1)

Best For:

Math reasoning
Code generation
Multi‑step chain‑of‑thought tasks
High‑complexity logic

3) Local Models (Ollama / LM Studio)

Pop supports local LLMs — ideal for offline and privacy‑sensitive scenarios.

Required Fields:

Local server address (default http://localhost:11434/)
Model name (e.g., llama3, qwen2.5, mistral)

Features:

Fully offline
Privacy‑safe
Speed depends on local hardware
Zero cost

Best For:

Enterprise intranet
Personal knowledge workflows
Document summarization and daily tasks

4) Enterprise / Custom API Models

For custom endpoints such as:

OpenAI‑compatible APIs
Custom embedding services
Enterprise gateways (ChatGLM, Qwen, DeepSeek private edition)

Required Fields:

API Key
Base URL (required)
Model name
Custom headers (optional)

Use Cases:

Corporate intranet deployment
Auditing & logging
Fully controlled model output

🧩 Model Parameters (Per‑Session Customization)

Each AI session can have its own model parameters:

Parameter	Description
temperature	Randomness — higher = more creative
top_p	Sampling range
presence_penalty	Reduce repetitive topics
frequency_penalty	Reduce repeated content
max_tokens	Maximum output length

Settings apply only to the current session.

🔄 Switching Models in a Conversation

Switching models is easy:

Open the Model Selector
Choose a model from the list
Takes effect immediately

Examples:

Window A uses DeepSeek
Window B uses GPT‑4o
Window C uses a local LLM

Perfect for comparing outputs.

🧪 Testing Model Connectivity

You can click Test Connection to:

Validate request success
Check response time
Confirm API Key correctness

If it fails, relevant error messages (401, 404, SSL, etc.) are shown.

⭐ Best Practices & Recommendations

Scenario	Recommended Model
Coding / Debugging	DeepSeek‑R1 / GPT‑4
Creative writing	GPT‑4o / GPT‑4o mini
Translation	GPT‑4o
Math reasoning	DeepSeek‑R1
Local privacy tasks	Ollama (Qwen / Llama)
Long‑document tasks	GPT‑4o / DeepSeek‑V3
Enterprise gateway	Custom OpenAI‑compatible API

📌 Summary

Pop’s multi‑model system allows users to:

Use the best model for each task
Switch models, roles, and knowledge bases freely
Combine cloud + local models
Meet both enterprise and personal AI needs