v1.78.0

Features

• feat(react-core)!: start Trajectory capture for every learning object (#7747) (d38d016)

Fixes

• fix(runtime): let COPILOTKIT_OPENAI_API choose the OpenAI API, and log the Chat Completions switch (refs PE-706) (#7754) (9158409)
• fix(runtime): keep slash model ids intact and use chat completions on OpenAI-compatible endpoints (refs PE-706) (#7726) (d1bdd92)
• fix(react-core,vue,angular): keep features enabled when the entitlement lookup times out (refs PE-533) (#7652) (be3a15c)
• fix(runtime): log a not-yet-created thread lookup at debug, not error (refs PE-678) (#7728) (c88c376)
• fix(runtime): default GoogleGenerativeAIAdapter to gemini-3.5-flash (#7716) (6b550dc)

Other Changes

• docs: recommend Gemini 3.x model ids new API keys can call (refs PE-690) (#7715) (4dc033a)

Breaking Changes

• fix(runtime): let COPILOTKIT_OPENAI_API choose the OpenAI API, and log the Chat Completions switch (refs PE-706) (#7754) (9158409)
  With OPENAI_BASE_URL set to a host other than OpenAI or Azure,
  "openai/..." model strings now use the Chat Completions API (#7726).
  Through a proxy in front of OpenAI (LiteLLM, Portkey, Helicone), models and
  options that only the Responses API supports, such as reasoning summaries,
  stop working. Set COPILOTKIT_OPENAI_API=responses to keep the Responses API.
• feat(react-core)!: start Trajectory capture for every learning object (#7747) (d38d016)
  A learning object without trajectoryId now starts capture after mount.
  To keep starting capture yourself with startTrajectory(), add
  autoStart: false, for example learning={{ onError, autoStart: false }}.