Skip to content

v2.10.0

v2.10.0: ctk 2.10.0
@queelius queelius tagged this 24 Apr 13:05 
BREAKING: consolidated LLM providers to a single OpenAI-compatible
implementation. Any endpoint speaking the OpenAI chat-completions
protocol works (real OpenAI, Azure, OpenRouter, vLLM, llama.cpp
server, LM Studio, muse, Ollama via /v1, etc.).

Highlights:
- Deleted ~2200 LOC of duplicated provider code (Ollama, Anthropic,
  OpenRouter, Local taggers and the Ollama/Anthropic LLM clients).
- Rewrote OpenAIProvider on the official openai SDK.
- New build_provider() factory honors ~/.ctk/config.json + flags + env.
- ctk tui enables chat by default (was opt-in). --no-chat disables.
- ctk chat no longer auto-connects to Ollama on localhost.
- providers.ollama / providers.anthropic removed from config defaults.
- tagging.default_provider: ollama -> openai.

Migration: if you had a local Ollama or Anthropic provider configured,
point providers.openai.base_url at your endpoint (Ollama: http://
localhost:11434/v1). Set OPENAI_API_KEY=unused for local servers
that do not enforce auth.

Tests: 2293 passing, 0 regressions. Net diff: +1041 / -3813 lines.
Assets 2
Loading