v1.1.0

Added

• Restructure as library + thin CLI binary (#153)
• (extends) Authenticated arai:extends for private rule sources (#154)
• (telemetry) Configurable endpoint for self-hosted collectors (#155)
• (audit) Arai audit --ship — send the audit trail to your own collector (#156)

Documentation

• (site) Surface v1.1 features and sharpen the value proposition (#158)

Fixed

• (grok) Realign integration with current Grok Build hooks contract (#159)

Refactored

• (lib) Item-level library API surface audit (#162) (#163)

Breaking changes (library API only)

The CLI surface, hook protocol, and on-disk formats are unchanged and
remain semver-stable. If you embed the arai library crate (new in
this release's lib/bin split), note that the pre-audit surface —
documented as unstable in 1.0.x — changed in the audit (#163):

• intent::Severity::from_str, intent::Timing::from_str, and
  intent::Action::from_str are renamed to from_str_lossy (they are
  infallible fallback parsers, not std::str::FromStr impls);
  Timing::as_str / Action::as_str now return &'static str
• Internal helpers left the public API (now pub(crate)):
  parser::extract_expiry / extract_noenrich / extract_annotations,
  discovery::parse_frontmatter, config::resolve_base_dir +
  ResolvedBaseDir / DeprecationNotice, Config hook-path helpers
  (claude_memory_*, claude_settings_path, grok_*_hooks_dir),
  Store::count_mcp_rules / query_tools_for_directory, and
  guardrails::sniff_content_for_tools_pub (wrapper removed)

From this release the pub items of the documented library modules are
item-level audited, documented, and covered by semver.

Fixed

• (grok) Realign the Grok Build (formerly Grok TUI / supergrok)
  integration with current docs.x.ai behavior: hook payload fields are
  now read in both snake_case (Claude protocol + Grok's Claude-compat
  layer) and camelCase (hookEventName/toolName/toolInput/
  sessionId, Grok Build native); .grok/hooks/arai.json registers
  only events Grok Build supports; arai init now tells users to run
  /hooks-trust (project hooks are trust-gated); tool aliases added
  for write_file/create_file/edit_file/apply_patch so
  Write/Edit-scoped rules can't silently no-fire on those names

Added

• Ārai now builds as a library crate (src/lib.rs) alongside the CLI
  binary, making the core — parser layers, rule store, guardrail
  matching, audit chain, hook decision logic — embeddable as a
  dependency. The CLI surface, hook contract, and DB schema are
  unchanged; src/main.rs is now a thin CLI over the library
  (#148)
• (extends) Authenticated arai:extends for private rule sources:
  arai trust --add <url> --bearer-env <ENV_VAR> attaches
  Authorization: Bearer <token> to fetches of that exact trusted URL
  (and its .sig sidecar). Only the variable name is stored; the token
  never reaches disk, logs, audit, telemetry, or error messages, and
  redirects stay disabled so the header cannot leak cross-host
  (#150)
• (telemetry) Configurable telemetry endpoint: [telemetry] endpoint
  (+ optional bearer_env) in config.toml points the existing queue at
  a self-hosted collector — same events, no PostHog api_key, 5 s
  timeout, queue retained on failure. Default sink unchanged when
  unset; ARAI_TELEMETRY=off / DO_NOT_TRACK=1 / enabled = false
  win regardless of endpoint (the config-level enabled = false
  opt-out is now actually enforced at the egress point — it was
  documented but unimplemented). Payload schema documented in
  docs/telemetry-payload.md
  (#151)
• (audit) arai audit --ship <url>: send audit day-buckets — plus
  their .head.YYYYMMDD chain sidecars — to your own HTTPS collector,
  so the hash chain can be verified server-side (tamper evidence
  survives transport). Resume cursor, idempotent re-ship (dedupe on
  per-entry chain hash), [ship] config section with optional
  bearer_env auth, non-zero exit on rejection, and never on the hook
  hot path. Collector reference: docs/audit-ship.md
  (#149)