Posts for: #How-To

Le Chonk Forgets: Replaying Mistral’s Thinking Through LiteLLM

Le Chonk Forgets: Replaying Mistral's Thinking Through LiteLLM

Mistral’s docs say to replay a reasoning model’s thinking on every turn, and warn that stripping it significantly degrades output quality. LiteLLM strips it. So does Qwen Code when it talks to Mistral directly. Here is how to tell whether your gateway is doing this to you, a third runtime patch that fixes it for Mistral without rebuilding anything, and what happened to a coding-agent run that had been going round in circles.

[Read more]

Le Chonk Meets LiteLLM: Two 422s and a Runtime Patch

Le Chonk Meets LiteLLM: Two 422s and a Runtime Patch

Mistral’s new Large 4 preview is picky about its request bodies, and LiteLLM sends it two things it refuses: its own MCP bookkeeping on replayed assistant turns, and tools with no parameters key. The upstream fix has been sitting unmerged since June. Here is a small runtime patch that fixes both without rebuilding the image, how to tell which of Mistral’s fifteen union errors is the real one, and what Le Chonk actually is.

[Read more]

Claude Code’s Own Sandbox, Inside Docker

Anthropic gave Claude Code its own bubblewrap sandbox. Turn it on inside a Docker container and it dies four layers deep – and the obvious fixes make it worse. Here is what is actually going on, and the two small profiles that make it work without handing the container CAP_SYS_ADMIN.

[Read more]