1.0.0b4

• Support for GPT-5 Codex. Still preliminary, but actually usable this time! Kudos to @sumeet
• Tracing to md files feature now not only writes streamed tokens in json format (.traces/*_RESPONSE_STREAM.md), but also the text that is being streamed in readable form (.traces/*_RESPONSE_TEXT.md).

1.0.0b3

• Added documentation for the LiteLLM Server Boilerplate part of this project (README_BOILERPLATE.md and librechat/README.md).
• Converted "tracing in markdown" into a generic feature that can be used by any LiteLLM Custom Provider (not only by the Claude Code GPT-5 Proxy, but also by the Yoda example and other future custom providers).
• Made the structures of the Claude Code GPT-5 Proxy (main branch) and LiteLLM Server Boilerplate (main-boilerplate branch) versions of the project more similar.
• Added docs/maintainers/CONVERT_TO_BOILERPLATE.md guide for maintainers.
• Converted docs/maintainers/DOCKER_PUBLISHING.md into docs/maintainers/RELEASE.md to document the release process from the beginning to the end.

1.0.0.1b3 (boilerplate release)

• Converted "tracing in markdown" into a generic feature that can be used by any LiteLLM Custom Provider (not only by the Claude Code GPT-5 Proxy, but also by the Yoda example and other future custom providers).
• Made the structures of the Claude Code GPT-5 Proxy (main branch) and LiteLLM Server Boilerplate (main-boilerplate branch) versions of the project more similar.

1.0.0b3-bp0 (boilerplate release)

• The very first my-litellm-server (LiteLLM server boilerplate) release - see README of main-boilerplate branch for details

1.0.0b2

• MAJOR REFACTORING: Project became dual purpose: Claude Code CLI proxy (original purpose) + LiteLLM server boilerplate (with LibreChat as the default Chat UI)
• The yoda_example package introduced with YodaSpeakLLM as an example of how to set up a custom LLM server using this repo as a "boilerplate"
• LibreChat was integrated to serve as a UI to communicate with Yoda example "pseudo-model" (or any other "pseudo-model" that could be set up in the "boilerplate") - see librechat folder
  • NOTE: Documentation is coming in future versions
• Issue #79 fixed (kudos to @spiochacz)

1.0.0b1

• REMAP_* env vars now map to GPT-5 by default, even when not set at all
• WRITE_TRACES_TO_FILES feature was introduced as a local alternative to Langfuse
• kill-docker.sh script was added
• WORK IN PROGRESS: Support for GPT-5-Codex (NOT RELIABLE YET: Tends to get stuck in loops due to incomplete API format conversion)

0.3.3

• Added ./uv-run.sh and ./run-docker.sh scripts for convenience
  (along with already existing ./deploy-docker.sh - see README for
  details).
• Documented (and supported in Docker) optional LITELLM_MASTER_KEY to
  enforce authentication at the level of the proxy (for remote
  deployments).
• Improved README.md and DOCKER_DEPLOYMENT.md overall.

0.3.2

• Docker support (Co-author: @amit-lavi)

0.3.1

• Highlighted error messages in red, so when something goes wrong, the actual problems are easier to spot in the long tracebacks.
• Made it possible to use models from any provider that LiteLLM supports, not only OpenAI and Anthropic (see the NOTE in "Available GPT-5 model aliases" section of README.md).
• Renamed OPENAI_ENFORCE_ONE_TOOL_CALL_PER_RESPONSE env var to ENFORCE_ONE_TOOL_CALL_PER_RESPONSE.

0.3.0

• Made Anthropic API key completely optional (made it possible to use OpenAI models as the "fast" model instead of Claude Haiku).
• Introduced REMAP_CLAUDE_HAIKU_TO, REMAP_CLAUDE_SONNET_TO, and REMAP_CLAUDE_OPUS_TO env vars to channel even Claude Code's built-in Agents (which don't inherit the model choice of the CLI) through OpenAI models.
• Fully resolved the problem of non-informative Exception without a message when something goes wrong with token streaming (#24).