Skip to content

0.4.0

Choose a tag to compare

View all tags
@bayger bayger released this 24 Apr 09:47
· 139 commits to release since this release
v0.4.0
f13f98c
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Breaking Changes

  • MASTER_ENCRYPTION_KEY is now required — the secrets subsystem is always active. The server will refuse to start if MASTER_ENCRYPTION_KEY is not set in the environment. Generate a 32-byte key with openssl rand -hex 32 and add it to your environment before upgrading. See the Secrets Management guide for details.

Added

  • Secrets management — a new first-class secrets subsystem that stores sensitive values encrypted at rest:
    • secrets database table with AES-256-GCM encryption via a MASTER_ENCRYPTION_KEY environment variable.
    • Provider configs and environment variables now accept opaque @sec:local:id references; they are resolved to plaintext automatically at runtime.
    • GET /api/secrets — list all secrets (values are never returned) with an orphans field identifying references that no longer appear in any config.
    • GET /api/secrets/:id/value — reveal the decrypted value for a specific secret (requires secrets:reveal permission).
    • DELETE /api/secrets/:id — delete a secret by ID.
    • New permissions: secrets:read, secrets:delete, secrets:reveal.
    • Migration script (npm run secrets:migrate) supports encrypting existing plaintext provider credentials in-place.
  • Analytics funnel engine — a new analytics sub-system for measuring sequential user journeys:
    • POST /api/projects/:projectId/analytics/funnels/query — execute an ad-hoc funnel query with up to 15 ordered steps. Supported event types: enter_stage, end_stage, action_fire, variable_changed, user_profile_changed, session_started, tool_response.
    • Response includes per-step userCount, percentage, dropoffCount, dropoffPercentage, and an overall totalConversionRate.
    • Relative and absolute time-range filtering (consistent with slice-and-dice analytics).
    • Saved Funnel Queries — persist and manage named funnel queries via CRUD endpoints under /api/projects/:projectId/analytics/funnels/saved-queries. Supports isShared flag and optimistic locking.
  • Extended analytics sources — three new data sources are now available in the slice-and-dice analytics query engine:
    • actions — action execution analytics from execution_plan events; one row per action per plan. Dimensions include actionName, stageName, and lifecycle context.
    • variables — one row per changed variable per variables_updated event.
    • user_profile — one row per changed field per user_profile_updated event.
    • The stage_visits source now includes stage names via a join, enabling stageName breakdowns.
  • Ollama LLM provider — built-in support for self-hosted Ollama instances:
    • New ollama provider type with configurable base URL.
    • Ollama models appear in the provider catalog with capability descriptions.
  • Project startingStageId — projects can now set a default starting stage; it is used as a fallback when the client omits stageId in the startConversation WebSocket message. Set to null to clear the default.

Improved

  • WebRTC audio pipeline — significant rework of the WebRTC channel's audio handling:
    • Native media tracks and RTCAudio interfaces replace the previous ad-hoc approach.
    • Outbound PCM audio is buffered into fixed-size frames and scheduled with real-time timing to reduce jitter and underruns.
    • Inbound audio format settings are managed per-connection rather than globally.

Fixed

  • Duplicate user messages were sent to the LLM during response generation — the current user input was included both via the conversation history (persisted before the LLM call) and explicitly appended, causing every response to see the user's last message twice.
Assets 2
Loading