Skip to content

v0.8.0

Choose a tag to compare

@github-actions github-actions released this 04 Jun 01:43
· 248 commits to main since this release
  • Changed: steer duplicate custom/synthetic corpora toward reuse
  • Changed: debounce re-gathers within a freshness window
  • Added: GET /metrics route for hosted HTTP deployments
  • Added: search_corpora tool for cross-corpus semantic search
  • Added: boot-time config validation for the HTTP serve path
  • Fixed: bind ETag cache path at load time so tests cannot clobber the real cache
  • Added: find_efforts topic-first effort discovery
  • Added: per-gather HTTP egress metrics on the datatracker load
  • Added: surface build version and cache freshness in /health
  • Added: deployment contract documentation for the HTTP serve path
  • Added: auto-update an installed Claude skill when it is pristine
  • Fixed: gather no longer aborts on PDFs with lone surrogates
  • Changed: rewrite corpus tool descriptions as routing rules vs web search
  • Changed: skip embedding concluded-draft revisions (rfc / replaced)
  • Changed: generalise email-domain-is-not-affiliation guidance
  • Added: opt-in MCP tools to start and monitor a corpus gather
  • Changed: fix WG-means-corpus conflations in SKILL.md and tool docstrings
  • Changed: use "corpus" terminology in user-facing CLI text
  • Changed: harden index build/read paths for concurrent serving
  • Added: remote OpenAI-compatible summariser backend
  • Fixed: parse the HTTP-date form of Retry-After in the remote embed backend
  • Added: structured JSON logging via IETF_LLM_LOG_FORMAT
  • Changed: resolve embed and summarise settings globally, not per-WG
  • Added: global service-config scope (env > CLI > global > default)
  • Fixed: typecheck with the optional embedding deps absent
  • Added: link GitHub authors to mailing-list identities via Datatracker
  • Added: GET /health readiness endpoint on the HTTP transport
  • Changed: make the server prewarm network-free for the remote backend
  • Added: Streamable HTTP transport for the MCP server
  • Added: configurable index directory for tmpfs placement
  • Added: record embedding dimension as index provenance
  • Fixed: keep schema_version across an index rebuild
  • Added: configurable cache and config directories via environment
  • Changed: query the embeddings index read-only
  • Changed: move NotebookLM export deps to the [notebooklm] extra
  • Changed: move on-device embeddings to the [local-embeddings] extra
  • Added: OpenAI-compatible remote embedding backend