v0.10.0-beta.14

Delta since previous beta (v0.10.0-beta.13) — 1 new changeset.

Minor Changes

• Add opt-in semantic search to the MCP search tool, powered by a remote OpenAI-compatible embeddings API. When enabled, an embeddings signal is fused into full_text ranking so an agent's query surfaces conceptually-related pages even when they share no keywords (e.g. "auth retries" → a page titled "Session Token Refresh"). It is additive and off by default: with the flag off, search is byte-for-byte today's lexical ranking, and the cmd-K omnibar always stays purely lexical and instant.

  Privacy / egress. When the feature is enabled and a key is set, the search query and matching page content are sent to the configured embeddings provider. Egress is explicit and scoped: only corpus content is embedded — anything excluded by .okignore / .gitignore is never sent — embedding is lazy (nothing leaves the machine until an agent actually runs a semantic search), and the key lives only in a 0600 ~/.ok/secrets.yml file (never in config, logs, or telemetry). If the key is missing, the provider errors, or you're offline, search silently degrades to lexical and never blocks.

  Setup (per machine): ok embeddings set-key stores the provider key in ~/.ok/secrets.yml (0600); set search.semantic.enabled: true in project-local config. ok embeddings status shows capability. Provider base URL / model / dimensions are configurable (Azure / self-hosted / other). Vectors cache incrementally under .ok/local/ (content-addressed, keyed by provider + model + dimensions). See the "Semantic search" section of the configuration reference for full setup.