yetone/magpie-releases v0.1.52 on GitHub

Features

Copilot models that come with their own terms (e.g. some Claude models) are now listed. magpie accepts the terms for your account the first time you use one, as VS Code does when you first pick it.
ChatGPT (Codex) account requests are now made the way Codex CLI makes them, so the newest models that answered "Model not found" to magpie work.

Claude subscription: a conversation's next turn continues in the same Claude Code session instead of starting over from the whole history, so its cached prompt is read again. New turns now write about 100 tokens to the cache instead of about 5,700; in tests, 79–87% of each conversation's input came from the cache, up from 37–63%.
ChatGPT (Codex) account: a conversation whose client sends no prompt_cache_key gets a stable one, raising cache hits from 74% to 96% in tests.
Copilot models no picker offers, that the account can't enable, or from Experimental vendors stay hidden.

Token counts reported to OpenAI- and Gemini-style clients for Claude now include prompt tokens written to the cache (Codex could show a 26K-token turn as 2 tokens).
Requests the agent hung up on after receiving the full answer (as Codex does) are no longer logged as canceled (499).

Download from usemagpie.ai, or curl -fsSL https://usemagpie.ai/install.sh | sh. An installed magpie updates itself.

macOS (Apple Silicon): magpie-darwin-arm64.dmg
macOS (Intel): magpie-darwin-amd64.dmg
Windows: magpie-windows-amd64.exe (ARM: magpie-windows-arm64.exe)
Linux: magpie-linux-amd64 / magpie-linux-arm64 (needs GTK 3 and WebKitGTK 4.1)
Terminal only: magpie-cli-<os>-<arch>