Highlights​

• New AI Flow Builder experience — visual flow definitions, versioning, templates, node editing, provider configuration, in-app testing, and runtime support.
• Shared @delphi/flow-engine package for execution, conversation timelines, runners, TTS/STT nodes, webhooks, conditions, and runtime command handling.
• Shared @delphi/features package with broader feature-flag coverage — registration, subscriptions, SMS, WebRTC, audio preprocessing, TTS media cache, AI Flow Builder, and QA scoring.
• Idempotent setup and seeding via a setup-step registry plus the new delphi-setup CLI for first-use and update flows.
• Expanded Vodafone TOBi support — improved LLM behavior, output hardening, usage metadata tracking, unique trace IDs, speaker logic, and managed TOBi flow handling.

Voice, calls, and WebRTC​

• Asterisk upgraded to 22.9.0 with missing RTP codec mappings filled in.
• Added chan_websocket / WebSocket-oriented voice infrastructure and improved TelPhi media-session handling.
• Fixed the WebRTC call-start UX that previously required two clicks.
• Improved endpoint selection, endpoint routing, proxy handling, WebSocket connectivity, and production/staging call reliability.
• Added SIP user-agent and transport handling improvements plus faster SIP header extraction.

Flow Builder and conversations​

• Flow definitions, versions, templates, and per-team flow briefings ship behind the aiFlowBuilder feature flag.
• Conversation/call detail views in TelWeb now expose Debug, Timeline, Flow run, QA, Token, and Action tabs.
• QA scoring is enqueued on flow finalization / hangup rather than ad hoc.

AI providers and models​

• Shared AI provider infrastructure: Azure, OpenAI, Pythia, Vodafone TOBi, TTS cache, proxy runtime, tracing, and a provider registry.
• TTS / speech text ingestion APIs with text normalization.
• Improved LLM key resolution and team WebRTC key handling.

API and SDK​

• New runtime capabilities and session-oriented TelAPI routes.
• Platform settings, platform secrets, flow briefing, flow templates, flow versions, and flow definition APIs.
• TelPhi SDK references and submodule versions bumped; see the SDK section for the current compatibility window.

Platform, billing, and onboarding​

• Registration flow: email verification ordering, legal-agreement UX, feature-flagged plan handling, auto-provisioning behavior.
• Billing activation, Stripe checkout/session handling, billing status APIs, and billing email reassignment.
• Subscription plan values refreshed; broader subscription/SMS guards across calls, APIs, tasker jobs, and usage tracking.
• The legacy hardcoded platform setup flow is replaced by the delphi-setup CLI.

Observability and debugging​

• Queryable SIP capture logging, trace IDs, call-ID correlation, multiline Kamailio parsing, improved SIP ladder / span persistence.
• Richer conversation and call detail views in TelWeb.
• Signoz integration and call observability tooling improvements.
• More idempotent TOBi conversation shutdown and better incomplete-span handling.

Infrastructure and operations​

• Docker production installs skip Husky prepare when dev dependencies are unavailable.
• New JANUS_WS_PORT env var for configuring the Janus WebSocket port. See Configuration.
• API proxy access and AWS Secret/LLM calls hardened behind proxy-aware paths.
• Improved container builds, TelWeb image setup, deployment scripts, service docs, and operational runbooks.
• TTS media-cache cleanup / check tooling and tasker support.

Fixes and polish​

• Fixed OpenAI TTS behavior.
• Fixed number syncing and base-number assignment paths.
• Fixed QA scoring enqueue behavior on flow finalization / hangup.
• Improved usage processing and display.

Upgrade notes​

SDK compatibility​

See the SDK section for installation and the per-version compatibility matrix.

Operator rollout checklist​

Everything in Upgrade notes above expands on these bullets. Work through them in order for a controlled bump:

1. Pull / roll artifacts — update ECR_TAG / CONFIG_REF and deploy using your normal playbook.
2. delphi-setup — TelWeb entrypoint applies Prisma migrations and baseline db seed only. After voiceai-telweb is healthy, run docker compose exec -it voiceai-telweb delphi-setup (first use / full registry) or delphi-setup --apply-missing on later image bumps. See First use on TelWeb.
3. JANUS_WS_PORT — required when you operate Janus yourself. Production setups often use 443 shared with HTTPS (WSS). See TelPro operations and the webrtc flag.
4. Proxy-aware AWS paths — confirm HTTP_PROXY, HTTPS_PROXY, and NO_PROXY on every private-network service that calls AWS Secrets Manager or AI providers after the hardened proxy routing in this release.
5. Feature flags — defaults keep new behavior off. Decide what to enable for each environment (aiFlowBuilder, qaScoring, audioPreprocessing, ttsMediaCache, registration, subscriptions, subscriptionManagement, sms, webrtc, apiAccess, additionalProviders, …). Follow Feature flags.
6. TelPhi SDK — consuming apps should stay inside >=0.1.0 <0.2.0; recommended v0.1.3. /sdk/compatibility.
7. TTS media-cache cleanup — Tasker ships periodic cleanup tooling; verify the scheduled job exists and Media / S3 / DB credentials resolve (Media).
8. Smoke-test WebRTC — place a browser call; the prior two-click start bug is fixed in this release.
9. Smoke-test carrier audio — TelSys is Asterisk 22.9.0 with RTP codec fixes; place an inbound call and confirm audio path.