AI Providers and Models
The Paquet Builder AI Assistant supports multiple AI providers — both cloud-based services and locally hosted models. This page helps you choose the right provider and model for your needs.
Cloud Providers
Section titled “Cloud Providers”| Provider | Recommended Model | Function Calling | Best For | Notes |
|---|---|---|---|---|
| OpenAI | gpt-5.4-mini | Excellent | Fast tasks, affordable usage | Reasoning effort option for o-series models |
| Anthropic (Claude) | claude-sonnet-4-6 | Excellent | Complex multi-step tasks | Best accuracy for tool calling |
| Anthropic (Claude) | claude-opus-4-6 | Excellent | Most demanding tasks | Premium, highest capability |
| Google (Gemini) | gemini-2.5-flash | Good | Budget-conscious usage | Free API tier available for low volume |
| Grok (xAI) | — | Good | Alternative provider | |
| Mistral | — | Good | European-hosted provider | EU data residency |
| DeepSeek | deepseek-chat | Good | Very affordable | Competitive pricing |
Local AI Providers
Section titled “Local AI Providers”Run AI models on your own machine — no API key, no internet, no data leaves your computer.
| Software | Provider Setting | Default Port | Notes |
|---|---|---|---|
| Ollama | Ollama | 11434 | Native integration, easiest setup |
| LM Studio | LlamaCpp | 1234 | User-friendly model browser |
| llama.cpp server | LlamaCpp | 8080 | Lightweight, for advanced users |
See Using Local AI Models for step-by-step setup instructions.
Choosing the Right Model
Section titled “Choosing the Right Model”Changing settings, adding files, querying project state — use a fast, affordable model:
- GPT 5.4-mini (OpenAI) — Best speed-to-quality ratio
- Gemini 2.5 Flash (Google) — Free tier, good for experimentation
Creating custom action workflows, configuring scope + signing + compression in one conversation:
- Claude Sonnet 4.6 (Anthropic) — Excellent tool-calling accuracy
- GPT 5.4 (OpenAI) — Strong multi-step reasoning
Full project setup from scratch, complex conditional custom actions, debugging build issues:
- Claude Opus 4.6 (Anthropic) — Highest overall capability
When data privacy is paramount or you want to avoid API costs:
- Ollama with Llama 3, Qwen 2.5, or Mistral — Requires a model with function calling support
Cost Considerations
Section titled “Cost Considerations”Usage costs depend on the provider and model you choose. Most cloud providers offer pay-as-you-go pricing based on the number of tokens (words) processed.
A typical Paquet Builder session with 10–20 exchanges costs:
- GPT 5.4-mini: A few cents
- Claude Sonnet 4.6: Around $0.05–0.15
- Claude Opus 4.6: Around $0.30–0.80
- Gemini 2.5 Flash: Free tier covers most usage
Provider-Specific Notes
Section titled “Provider-Specific Notes”OpenAI
Section titled “OpenAI”- Reasoning Effort: Available for o-series and GPT-5 models. Controls how much “thinking” the model does before responding. Higher effort = better results for complex tasks but slower and more expensive.
- The default model (when the field is empty) is
gpt-4o.
Anthropic (Claude)
Section titled “Anthropic (Claude)”- Generally provides the most reliable tool-calling behavior for multi-step operations.
- Supports both Sonnet (faster, cheaper) and Opus (more capable) tiers.
Google (Gemini)
Section titled “Google (Gemini)”- Free API tier available at aistudio.google.com with generous usage limits.
- Some complex tool-calling scenarios may require retry (handled automatically by Paquet Builder).
DeepSeek
Section titled “DeepSeek”- Very competitive pricing, often 10-20x cheaper than equivalent Western providers.
- Good function calling support for most operations.