Skip to content

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.1.0] - 2026-06-04

First public PyPI release.

Distributed on PyPI as ollama-proxy-plus (the ollama-proxy name was already taken). The Python import path remains ollama_proxy and the CLI command is ollama-proxy.

Added

  • CLI entry point: ollama-proxy command with serve (default), doctor, list-models, and test subcommands.
  • YAML config file support via --config flag or PROXY_CONFIG_FILE env var.
  • Auto-discovery of OpenAI-compatible providers from PROVIDERS_<NAME>_API_KEY / PROVIDERS_<NAME>_BASE_URL env vars.
  • Per-provider retry and fallback configuration in YAML config.
  • Health endpoints: /health (basic) and /health/providers (connectivity).
  • Structured logging with per-request correlation IDs (x-request-id header).
  • JSON log format option via PROXY_LOG_JSON=1.
  • EXTRA_MODELS env var to add models to /api/tags without editing config.
  • AWS Bedrock provider via boto3 Converse API.
  • OpenAI-compatible endpoint at /v1/chat/completions with proper tool calls and streaming SSE support (for GitHub Copilot, Continue, Cursor).
  • tools and format (JSON mode / JSON schema) pass-through on /api/chat.
  • Optional dependency groups: [anthropic], [bedrock], [all].

Changed

  • Gemini now routes through Google's OpenAI-compatible endpoint instead of the deprecated google-generativeai SDK.
  • Kimi now uses api.moonshot.ai (was .cn) and supports thinking-mode control via extra_body.
  • /api/show accepts any prefix/model-id whose prefix is configured (registry membership is no longer required for direct API calls).
  • Reported Ollama version bumped to 0.6.5 (some clients require ≥0.6.4).

Removed

  • google-generativeai dependency (Gemini handled via OpenAI-compat path).
  • Standalone OpenAIProvider and KimiProvider classes (folded into OpenAICompatProvider for consistency).