v0.30.0

Added

• Embedding fallback replica: Configure one compatible secondary endpoint for queries, documents and indexing batches. Calls prefer the primary and switch on network failures, timeouts or HTTP 401/403/404/429/5xx, including when the primary sends an eligible status and then drops the connection while the error body is read. Supports OpenAI-compatible and native Ollama protocol changes, Google-compatible replicas, offline primary startup, independent credentials and project-level disabling without rebuilding compatible indexes. An Ollama replica keeps its default request-size guards, so a replayed batch is split into requests of at most sixteen texts.

Fixed

• Ollama evaluation gate: Corrected eval:ci:ollama to use the documented absolute Ollama budget without comparing against the paid-provider baseline.
• reStructuredText discovery: .rst documentation files are now included by default and use the existing bounded generic-text parser.
• Identifier-bearing explanatory queries: Only broad conceptual/overview cues now keep the conceptual ranking path. Queries that are identifier-focused and only say “explain”/“explanation” now stay source-oriented/neutral, preserving explicit definition, implementation, documentation, test, configuration, and call-flow priorities.
• CLI output redaction: Successful cbi index output now redacts API keys, tokens, passwords, secrets, and authorization values before writing to stdout.

Published as open-codebase-index@0.30.0 and opencode-codebase-index@0.30.0.