Model Picker - Perplexity

2026-07-06 · Source: perplexity.ai via Google News · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, medium

Summary

Perplexity's Model Picker is a command-line example demonstrating how to use the Perplexity Agent API to generate grounded recommendations for open models from Hugging Face. It utilizes a single Agent API request that integrates two distinct tools: the "mcp" tool, which connects to the Hugging Face MCP server for live model registry data (e.g., download counts, licenses, last-modified dates), and the "web_search" tool, used for checking benchmarks, quality, and known issues. Users provide a plain-language task, such as "on-device English speech-to-text, small enough to run on a laptop," and the system returns a Markdown brief containing a shortlist of models and a justified recommendation. The Agent API is explicitly instructed to verify all claims using these tools, ensuring recommendations are based on current data rather than the model's training memory. Installation requires the Perplexity Python SDK, with an optional "HF_TOKEN" for increased Hugging Face rate limits. The process involves the model iteratively searching the Hub, then performing web searches before presenting its final output.

Key takeaway

For AI Engineers evaluating open-source models for specific deployment constraints, Perplexity's Model Picker offers a robust, data-driven approach. Instead of relying on a model's potentially outdated training data, you can use this tool to get real-time, verified recommendations grounded in live Hugging Face registry data and web benchmarks. This ensures your model choices are current and optimized for factors like on-device performance or license compatibility, streamlining your selection process and reducing manual cross-referencing.

Key insights

Combining live registry data with web search grounds LLM model recommendations.

Principles

• Ground LLM outputs in real-time data.
• Combine specialized tools for comprehensive results.
• Explicitly instruct models to use tools.

Method

The Agent API uses "mcp" for Hugging Face registry search and "web_search" for external validation. It iteratively refines searches, retrieves details, and then performs web checks before generating a recommendation.

In practice

• Use Perplexity Agent API for model selection.
• Integrate "mcp" and "web_search" tools.
• Set "HF_TOKEN" for higher rate limits.

Topics

• Perplexity Agent API
• Hugging Face MCP
• Model Recommendation
• Open-source Models
• Speech-to-Text
• Command-line Tools

Code references

• ppl-ai/perplexity-python

Best for: AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• Perplexity with LiveKit Agents - Perplexity — The `livekit-plugins-perplexity` package enables LiveKit Agents to use Perplexity's web-grounded LLMs.
• Perplexity with AG2 - Perplexity — The AG2 framework now integrates Perplexity Search API for direct, token-efficient web search capabilities within multi-agent systems.
• Windsurf 1.9566.9 — Model selection and platform stability are enhanced through new UI features and backend improvements.
• Perplexity Computer vs OpenClaw 2026.3.1 — Choosing the correct AI agent for a specific task is crucial to prevent operational failures and inefficiencies.
• Model Organisms Are Leaky: Perplexity Differencing Often Reveals Finetuning Objectives — Perplexity differencing reveals finetuning objectives in LLMs by exploiting their overgeneralization of learned behaviors.
• Perplexity's 19-model AI ‘Computer’ — Perplexity Computer orchestrates 19 AI models for autonomous, long-running, sandboxed agentic task execution.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by perplexity.ai via Google News.