🛠️Simon Willison's LLM 0.32 Turns the CLI Into an Agent
TL;DR
Simon Willison shipped LLM 0.32 on August 4, his biggest release since the tool launched. It adds visible reasoning traces, server-side provider tools like code execution and web search, and a Git-style content-addressable SQLite log.
Simon Willison shipped LLM 0.32 on August 4, his biggest release since the tool launched. It adds visible reasoning traces, server-side provider tools like code execution and web search, and a Git-style content-addressable SQLite log. He now calls LLM an agent framework.
Key Points
Reasoning models now stream their traces to standard error
Server-side tools: OpenAI code interpreter, web search, Anthropic MCP
New content-addressable message store modeled after Git
Default model is now the inexpensive GPT-5.6 Luna
Why It Matters
A single CLI that mixes models, tools, and MCP servers in one line is quietly becoming a practical agent framework for developers who live in the terminal.
Quick Facts
Frequently Asked Questions
Why does this matter?
A single CLI that mixes models, tools, and MCP servers in one line is quietly becoming a practical agent framework for developers who live in the terminal.
What happened?
Simon Willison shipped LLM 0.32 on August 4, his biggest release since the tool launched. It adds visible reasoning traces, server-side provider tools like code execution and web search, and a Git-style content-addressable SQLite log.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 2,647 builders reading daily.