v1.1.71

Added

• FastContext Multi-Provider Support via LiteLLM: The Python runner now uses LiteLLM as a unified adapter to support OpenAI, Anthropic, OpenRouter, and custom providers. Falls back to the native OpenAI SDK if LiteLLM is not installed.
• LiteLLM Dependency in Setup Scripts: Both \setup-fastcontext.ps1\ (Windows) and \setup-fastcontext.sh\ (Linux/macOS) now install and verify \litellm>=1.74.0\ alongside existing dependencies.
• Provider-Aware Fallback Models: Credential resolver now returns \providerType, \providerName, \ ierName, and \providerMismatch\ metadata, and uses \DEFAULT_FALLBACK_MODELS\ to pick sensible default models per provider type (OpenAI → \gpt-4o, Anthropic → \claude-sonnet-4-20250514, OpenRouter → \�nthropic/claude-sonnet-4-20250514).
• Unique Trajectory Paths: Each FastContext invocation generates a unique trajectory JSONL file (\ rajectory--.jsonl) in .fastcontext/, preventing collisions during concurrent runs. Stale trajectory files are automatically cleaned up.
• Live Model/Provider Info in Logs: FastContext displays the resolved model name, tier, provider name, and provider type at the start of each run, with a warning if the tier's configured provider was not found.
• Backend Info in Start Events: The \start\ event in live logging now includes a \�ackend\ field indicating whether LiteLLM or the native OpenAI SDK is being used.

Changed

• Improved Tier Resolution Logic: Credential resolver now explicitly checks
  esearcher.model, \subagentDefault.model, and falls back to the main tier (\master/\superagent), with clear tier name tracking and provider mismatch detection.
• CLI Args Extended: FastContext runner now accepts --trajectory-path\ and --provider\ flags.
• Tool Description Updated: FastContext tool description now accurately reflects the tier resolution order:
  esearcher > subagentDefault > main fallback.