v1.8.0

[ldyrmishiritech]

v1.8.0

This release is a major expansion of the platform: enterprise single sign-on with Microsoft Entra ID, an in-app notification system with email delivery, a built-in Help Center backed by Azure DevOps Boards, real-time voice conversations and pluggable audio providers, a visual workflow builder with cron-based scheduling, a versioned prompt editor, local model fine-tuning, GDPR-compliant conversation deletion, and direct-to-S3 file uploads. It also brings significant security hardening (PII masking, safe model loading, refresh-token rotation, OAuth2 for MCP) and a reworked Celery worker topology for safer, more scalable background processing.

Authentication & SSO

• Added Microsoft Entra ID single sign-on (OIDC authorization-code + PKCE flow) with automatic first-login user provisioning, linked on the entra_oid claim, and configurable tenant scope (specific tenant or common/organizations/consumers).
• Introduced conversation-scoped authentication and permissions — guest JWTs, token-based agent access, and tightened public-embed API-key handling with operator-level authorization checks.
• Hardened embed / public-API security (RSA public/private key authentication, agent security dependencies).
• Implemented refresh-token rotation: /auth/refresh now validates the token type and revokes the old refresh token after use.

Notifications

• Added an in-app notification system with typed notifications (info, success, warning, error), a WebSocket-backed feed, unread counters, and read/unread state.
• User and group targeting — admins configure which users or groups receive each notification type; notifications auto-scope to a user's group membership and supervised groups.
• Added an email delivery channel via SMTP (master switch EMAIL_ENABLED, configurable sender/host/TLS), so notifications can be dispatched by email.
• Per-user preference overrides let users suppress notification types they don't want.

Help Center & Support Tickets

• Added a Help Center / support-ticket system — create, list, comment on, and track tickets, with duplicate-candidate detection to reduce noise.
• Azure DevOps Boards integration — tickets sync to Azure DevOps work items (Bug / Feature / Task) with bi-directional status and comment updates; configurable area path and per-type work-item overrides.
• Inbound Azure DevOps webhook keeps work-item state in sync back into GenAssist (optional shared-secret verification).
• Added reported-feedback / message-issue tracking — flagged messages move through an open → resolved → dismissed workflow.

Voice & Audio

• Added real-time voice conversations backed by the Gemini Live audio API (live speech-to-speech), with automatic transcription and conversation-memory persistence.
• New workflow nodes: STT (speech-to-text), TTS (text-to-speech), and Voice Agent (STT → agent → TTS).
• Pluggable audio providers — Gemini, ElevenLabs, Google Cloud, and OpenAI, selectable per workflow node and managed centrally.
• Transcript messages can now store and stream message audio.

Workflows

• Added a visual workflow builder API that creates workflows from a structured JSON spec (branching, tool connections, per-node config overrides) with automatic layout.
• Added workflow scheduling — cron-based recurring runs with run-history tracking and on-demand "run now"; backed by Celery scheduled-run and stuck-run reconciliation tasks.
• New runtime nodes: Finalize Conversation, External Agent (call external REST APIs), File Reader (extract text from uploaded files), and Create Workflow Schedule.
• Added MCP OAuth2 support — MCP servers can authenticate via OAuth2 / OIDC with JWKS discovery and token validation, alongside static API keys.

Prompt Editor

• Added a version-controlled prompt editor for workflow LLM nodes — create, restore, and delete prompt versions per node/field.
• Added prompt evaluation and optimization at the node level, with gold-suite linking for repeatable testing.

Analytics & Conversations

• Reworked the analytics dashboard with KPI metrics, attribute breakdowns, persistent/comparison date ranges, and custom-attribute filtering.
• Added group scoping across analytics and conversations (group_id), plus an analytics backfill task.
• Added supervisor indicators and improved supervisor takeover handling in conversations.

Local Fine-Tuning (BETA)

• Added a local fine-tuning proxy that forwards training, deployment, and inference requests to an external fine-tune service using short-lived minted JWTs.
• Manage jobs, query available GPUs and supported models, and deploy/test fine-tuned models from the UI without exposing the upstream service.

File Management

• Added direct browser → S3 presigned PUT uploads (opt-in) to bypass the API for large files and reduce server I/O.
• Added chunked / resumable upload sessions for both the file manager and the knowledge base.
• Raised and made configurable the upload-size limits (request body, knowledge base, and file manager).

Privacy & Compliance

• Added GDPR conversation deletion with soft, anonymize, and hard delete modes (admin-gated); deletion also purges associated S3/cloud file objects.
• Added PII / sensitive-data masking — heuristic detection and redaction of passwords, API keys, tokens, emails, SSNs, credit cards, phone numbers, and JWTs in logs and before workflow execution.
• Added a restricted "safe pickle" loader for ML models (allows only known-safe classes) to prevent code-execution exploits.
• Added data-residency enforcement — LLM requests to providers outside a configured residency zone are rejected.

Frontend / UX

• New views: Analytics (refactored), Help Center, Reported Feedback, Notifications, Audio Providers, Local Fine-Tune, GDPR Conversations, Settings, and SSO login.
• New workflow Scheduling and Prompt Editor views in the AI Agents workflow builder.
• Added skeleton loading states and improved empty-state components across pages and tables.

Chat Widget (plugins)

• React widget: added live voice calls (LiveCallControl, useLiveVoice), audio playback (AudioPlayer), file attachments (useFileAttachments), a customizable chat-bubble icon, and refactored composition hooks (scroll, viewport, thinking animation, theming).
• JavaScript widget: added BUILD.md build guide, switched to a gitignored font.example.css template, and simplified the Vite IIFE bundle workflow.

Dependencies & Security

• Backend: security/maintenance bumps — cryptography 46.0.5→46.0.7, urllib3 2.6.3→2.7.0, PyJWT[crypto] 2.10.1→2.12.0, lxml 5.4.0→6.1.0, pypdf 6.6.2→6.10.2, langchain-core 1.2.7→1.3.3, langsmith 0.5.0→0.8.0, plus new PyOpenSSL>=26.0.0. New libraries: aiosmtplib (email), litellm, google-genai (Gemini Live). Removed langchain-chroma.
• Frontend: vite 5→6, react-router-dom 6.26→6.30, axios 1.15→1.16, genassist-chat-react 1.0.37→1.0.39, plus lodash, postcss, and uuid bumps.

---

New Endpoints

Audio Providers (/audio-providers)

Method	Path	Params	Example
GET	/audio-providers	—	GET /audio-providers
GET	/audio-providers/minimal	—	GET /audio-providers/minimal
GET	/audio-providers/by-capability/{capability}	capability (path: tts|stt|both)	GET /audio-providers/by-capability/tts
GET	/audio-providers/form-schemas	—	GET /audio-providers/form-schemas
GET	/audio-providers/node-schemas	—	GET /audio-providers/node-schemas
GET	/audio-providers/{provider_id}	provider_id (path, UUID)	GET /audio-providers/3fa85f64-…
POST	/audio-providers	Body: AudioProviderCreate (name, type, config)	POST /audio-providers
PATCH	/audio-providers/{provider_id}	provider_id (path, UUID); Body: AudioProviderUpdate	PATCH /audio-providers/3fa85f64-…
DELETE	/audio-providers/{provider_id}	provider_id (path, UUID)	DELETE /audio-providers/3fa85f64-…
POST	/audio-providers/test-connection	provider_id (query, UUID, optional); Body: AudioProviderBase (type, capability, connection_data)	POST /audio-providers/test-connection

Microsoft Entra SSO (/auth/sso/microsoft)

Method	Path	Params	Example
GET	/auth/sso/microsoft/status	— (public — is SSO enabled)	GET /auth/sso/microsoft/status
GET	/auth/sso/microsoft/start	— (redirects to Microsoft; rate-limited)	GET /auth/sso/microsoft/start
GET	/auth/sso/microsoft/callback	code (query), state (query)	GET /auth/sso/microsoft/callback?code=…&state=…
POST	/auth/sso/microsoft/complete	Body: MicrosoftSsoCompleteRequest (code: str, 8–256)	POST /auth/sso/microsoft/complete

Azure DevOps Webhook (/integrations/azure-devops)

Method	Path	Params	Example
POST	/integrations/azure-devops/webhook	Header X-ADO-Webhook-Secret (optional); Body: AdoWebhookPayload (resource, eventType, …)	POST /integrations/azure-devops/webhook

Local Fine-Tuning (/local-fine-tuning)

Method	Path	Params	Example
GET	/local-fine-tuning/supported-models	skip (query, int, default 0), limit (query, int, default 10)	GET /local-fine-tuning/supported-models
GET	/local-fine-tuning/system/gpus	—	GET /local-fine-tuning/system/gpus
GET	/local-fine-tuning/jobs	—	GET /local-fine-tuning/jobs
GET	/local-fine-tuning/jobs/{job_id}	job_id (path)	GET /local-fine-tuning/jobs/abc123
GET	/local-fine-tuning/jobs/{job_id}/events	job_id (path)	GET /local-fine-tuning/jobs/abc123/events
POST	/local-fine-tuning/jobs	Body: CreateLocalFineTuneJobRequest	POST /local-fine-tuning/jobs
POST	/local-fine-tuning/jobs/{job_id}/cancel	job_id (path)	POST /local-fine-tuning/jobs/abc123/cancel
DELETE	/local-fine-tuning/jobs/{job_id}/files	job_id (path); delete_data_files (query, bool, default true), delete_checkpoints (query, bool, default true), delete_model (query, bool, default false)	DELETE /local-fine-tuning/jobs/abc123/files
GET	/local-fine-tuning/deployments	—	GET /local-fine-tuning/deployments
GET	/local-fine-tuning/deployments/{deployment_id}	deployment_id (path)	GET /local-fine-tuning/deployments/dep123
POST	/local-fine-tuning/deployments	Body: CreateDeploymentRequest	POST /local-fine-tuning/deployments
DELETE	/local-fine-tuning/deployments/{deployment_id}	deployment_id (path)	DELETE /local-fine-tuning/deployments/dep123
GET	/local-fine-tuning/deployments/{deployment_id}/health	deployment_id (path)	GET /local-fine-tuning/deployments/dep123/health
POST	/local-fine-tuning/deployments/{deployment_id}/test-inference	deployment_id (path); Body: TestInferenceRequest (message: str)	POST /local-fine-tuning/deployments/dep123/test-inference

Notifications (/notifications)

Method	Path	Params	Example
GET	/notifications	— (current user's settings)	GET /notifications
PATCH	/notifications	Body: NotificationUserSettingsUpdate	PATCH /notifications
GET	/notifications/admin/targeting	—	GET /notifications/admin/targeting
PUT	/notifications/admin/targeting/{type_key}	type_key (path); Body: NotificationTypeTargetingUpdate	PUT /notifications/admin/targeting/conversation_started
GET	/notifications/state/counters	— (unread counters)	GET /notifications/state/counters
PATCH	/notifications/state/read	Body: NotificationStateUpdate (notification_ids: UUID[], is_read: bool, optional)	PATCH /notifications/state/read

Prompt Editor (/genagent/prompt-editor)

Method	Path	Params	Example
GET	/genagent/prompt-editor/versions/{workflow_id}/{node_id}/{prompt_field}	all path	list versions
POST	/genagent/prompt-editor/versions/{workflow_id}/{node_id}/{prompt_field}	path; Body: PromptVersionCreate	create version
POST	/genagent/prompt-editor/versions/{version_id}/restore	version_id (path, UUID)	restore version
DELETE	/genagent/prompt-editor/versions/{version_id}	version_id (path, UUID)	soft-delete version
DELETE	/genagent/prompt-editor/versions/{version_id}/hard	version_id (path, UUID)	permanently delete
GET	/genagent/prompt-editor/config/{workflow_id}/{node_id}/{prompt_field}	all path	get config
PUT	/genagent/prompt-editor/config/{workflow_id}/{node_id}/{prompt_field}/gold-suite	path; Body: GoldSuiteLinkRequest (suite_id: UUID, name: str)	link gold suite
POST	/genagent/prompt-editor/evaluate/{workflow_id}/{node_id}/{prompt_field}	path; Body: PromptEvalRequest (prompt_content, techniques, provider_id)	evaluate prompt
POST	/genagent/prompt-editor/optimize/{workflow_id}/{node_id}/{prompt_field}	path; Body: PromptOptimizeRequest (current_prompt, provider_id, instructions, failed_cases)	optimize prompt

Support Tickets (/help-center/tickets)

Method	Path	Params	Example
GET	/help-center/tickets/duplicates	title (query, min 3), ticket_type (query, default bug)	GET /help-center/tickets/duplicates?title=…
GET	/help-center/tickets	status, ticket_type, skip (default 0), limit (default 50, max 100), mine_only (default false) — all query	GET /help-center/tickets
POST	/help-center/tickets	Body: SupportTicketCreate	POST /help-center/tickets
GET	/help-center/tickets/{ticket_id}	ticket_id (path, UUID)	GET /help-center/tickets/3fa85f64-…
GET	/help-center/tickets/{ticket_id}/comments	ticket_id (path, UUID)	list comments
POST	/help-center/tickets/{ticket_id}/comments	ticket_id (path, UUID); Body: SupportTicketCommentCreate	add comment

Voice Live (/voice)

Method	Path	Params	Example
WebSocket	/voice/live/{agent_id}	agent_id (path), thread_id (query, required). Client sends binary audio frames; server returns binary audio + JSON transcript/error events.	WS /voice/live/3fa85f64-…?thread_id=…

Workflow Builder (/workflow-builder)

Method	Path	Params	Example
POST	/workflow-builder/config/from-builder	Form body: workflow_name (str, required), workflow_json (str, required — serialized builder spec), workflow_description (str, optional)	POST /workflow-builder/config/from-builder

Workflow Schedules (/genagent/workflow-schedules)

Method	Path	Params	Example
GET	/genagent/workflow-schedules	is_active (query, bool, optional)	GET /genagent/workflow-schedules
POST	/genagent/workflow-schedules	Body: WorkflowScheduleCreate (workflow_id, cron_expression, is_active, …)	POST /genagent/workflow-schedules
GET	/genagent/workflow-schedules/{schedule_id}	schedule_id (path, UUID)	GET /genagent/workflow-schedules/3fa85f64-…
PUT	/genagent/workflow-schedules/{schedule_id}	schedule_id (path, UUID); Body: WorkflowScheduleUpdate	PUT /genagent/workflow-schedules/3fa85f64-…
DELETE	/genagent/workflow-schedules/{schedule_id}	schedule_id (path, UUID)	soft-delete
GET	/genagent/workflow-schedules/{schedule_id}/runs	schedule_id (path); run_status (query, enum, optional), limit (default 50), offset (default 0)	run history
POST	/genagent/workflow-schedules/{schedule_id}/run-now	schedule_id (path, UUID)	trigger immediate run

---

Updated Endpoints

Analytics (/analytics)

Method	Path	Change
POST	/analytics/backfill	New — trigger analytics backfill (from_date, to_date query, optional)
GET	/analytics/groups/{group_id}/agents	New — list agents for a group
GET	/analytics/* (daily-stats, summary, agent-stats, node-stats, audio metrics, custom-attribute keys/breakdown, exports)	All stats endpoints now accept optional group_id (query, UUID) for group-scoped filtering

API Keys (/api-keys)

Method	Path	Change
POST	/api-keys/{api_key_id}/reveal	New — decrypt/reveal a masked key (perm ApiKey.DECRYPT)

Conversations (/conversations)

Method	Path	Change
GET	/conversations/issues	New — list reported message issues (skip, limit, status)
GET	/conversations/{conversation_id}/messages/{message_id}/audio	New — stream message audio
PATCH	/conversations/issues/{message_feedback_id}/status	New — update issue resolution status
DELETE	/conversations/{conversation_id}/gdpr	New — GDPR erasure; mode (query: soft|anonymize|hard); admin + Conversation.DELETE_GDPR
PUT	/conversations/in-progress/{conversation_id}	Now accepts multipart/form-data with model (JSON), optional audio_file, and audio_format (default webm)

File Manager (/file-manager)

Method	Path	Change
POST	/file-manager/upload	New — bulk upload convenience endpoint
POST	/file-manager/upload-session	New — create chunked upload session
POST	/file-manager/upload-session/{session_id}/chunk	New — append chunk (chunk_index)
POST	/file-manager/upload-session/{session_id}/complete	New — finalize chunked upload
GET	/file-manager/upload-session/{session_id}	New — session status
GET	/file-manager/{file_id}, /file-manager/source/{file_id}	New force_proxy (query, bool, default false) — S3-backed files redirect to a presigned URL unless force_proxy=true

Agent Knowledge (/genagent/knowledge)

Method	Path	Change
POST	/genagent/knowledge/upload-session (+ /chunk, /complete, GET .../{session_id})	New — chunked KB upload sessions
POST	/genagent/knowledge/upload	Validates against KNOWLEDGE_MAX_UPLOAD_BYTES; supports S3 storage provider

Other

Method	Path	Change
POST	/genagent/workflow/{workflow_id}/execute, /genagent/workflow/test	Apply PII redaction (credit card, IBAN, SSN) to the message input before execution; new supported node types in the workflow body: fileReaderNode, externalAgentNode, ttsNode, sttNode, voiceAgentNode, createWorkflowScheduleNode
POST	/auth/refresh	Body refresh token now must be typ=refresh (access tokens rejected) and is rotated (old token revoked)
GET	/dashboard/notifications	New — paginated notification feed with type/level filters and unread count
POST	/internal/ws-token/verify	Body now accepts both guest (conversation-scoped) and user tokens

---

Environment Variable Changes

Backend (backend/.env.example + settings.py) — Added

The Required? column indicates what you must set for a fresh production deployment. "Yes (prod)" = the app fails fast without it; "If <feature>" = only required when that feature is turned on; "No" = has a safe default.

Required for production

Variable	Required?	Default	Purpose
SEED_ADMIN_PASSWORD	Yes (prod)	change-me (dev only)	Admin seed password — seeding raises in non-dev if unset; also required to create a tenant
SEED_APIUSER_PASSWORD	Yes (prod)	change-me (dev only)	API-user seed password — required to create a tenant
SEED_SUPERVISOR_PASSWORD	Yes (prod)	change-me (dev only)	Supervisor seed password — required when that user is seeded
SEED_OPERATOR_PASSWORD	Yes (prod)	change-me (dev only)	Operator seed password — required when that user is seeded
SEED_TRANSCRIBE_OPERATOR_PASSWORD	Yes (prod)	change-me (dev only)	Transcribe-operator seed password — required when that user is seeded
SEED_ADMIN_USERNAME	No	admin	Admin seed username
SEED_SUPERVISOR_USERNAME	No	supervisor1	Supervisor seed username
SEED_OPERATOR_USERNAME	No	operator1	Operator seed username
SEED_APIUSER_USERNAME	No	apiuser1	API-user seed username
SEED_TRANSCRIBE_OPERATOR_USERNAME	No	transcribeoperator1	Transcribe-operator seed username

In dev/debug the seed passwords fall back to change-me. In production (DEBUG/DEV/FASTAPI_DEBUG all false) a missing seed password raises ValueError and aborts seeding / tenant creation.

Microsoft Entra ID SSO — required only when SSO_MICROSOFT_ENABLED=true

Variable	Required?	Default	Purpose
SSO_MICROSOFT_ENABLED	No	false	Master switch for Entra ID SSO
SSO_MICROSOFT_ENTRA_TENANT_ID	If SSO	—	Entra directory (tenant) ID
SSO_MICROSOFT_CLIENT_ID / SSO_MICROSOFT_CLIENT_SECRET	If SSO	—	App registration credentials
SSO_MICROSOFT_REDIRECT_URI	If SSO	—	Backend OAuth callback (must match Entra "Web" redirect, includes /api)
SSO_MICROSOFT_POST_LOGIN_FRONTEND_URL	If SSO	—	SPA origin used after login
SSO_MICROSOFT_POST_LOGIN_ORIGINS_ALLOWLIST	No	—	Extra allowed post-login origins (CSV)
SSO_MICROSOFT_AUTO_PROVISION	No	false	Create local user on first Entra login
SSO_MICROSOFT_DEFAULT_USER_TYPE_ID / SSO_MICROSOFT_DEFAULT_ROLE_IDS	If auto-provision	—	Defaults applied when auto-provisioning

Other feature-gated / optional

Variable	Required?	Default	Purpose
AZURE_DEVOPS_ORGANIZATION_URL	If ADO sync	https://dev.azure.com/your-org	Help Center → Azure Boards org
AZURE_DEVOPS_PROJECT	If ADO sync	—	Target project
AZURE_DEVOPS_PAT	If ADO sync	—	Personal access token
AZURE_DEVOPS_WORK_ITEM_TYPE	No	Bug	Work-item type for bug tickets
AZURE_DEVOPS_FEATURE_WORK_ITEM_TYPE / AZURE_DEVOPS_TASK_WORK_ITEM_TYPE	No	—	Overrides for feature/task types
AZURE_DEVOPS_DEFAULT_AREA_PATH	No	—	Area path for client-reported work items
AZURE_DEVOPS_WEBHOOK_SECRET	No	—	Shared secret for the inbound webhook
HELP_CENTER_PUBLIC_BASE_URL	No	—	Public URL embedded in ADO work-item links
EMAIL_ENABLED	No	true	Master switch for SMTP email (logs instead of sends when false)
SMTP_HOST / SMTP_PORT	If email sends	— / 587	SMTP server (587 STARTTLS, 465 implicit TLS)
SMTP_USER / SMTP_PASSWORD	If email sends	—	SMTP auth
SMTP_FROM_EMAIL / SMTP_FROM_NAME	If email sends	— / GenAssist	Sender identity
SMTP_USE_TLS / SMTP_TIMEOUT	No	true / 15	STARTTLS toggle / timeout (s)
FILES_DIRECT_S3_UPLOAD_ENABLED	No	false	Opt-in browser → S3 presigned PUT uploads (requires FILE_MANAGER_PROVIDER=s3)
FILES_DIRECT_S3_PRESIGN_EXPIRES_SECONDS	No	600	Presigned PUT URL lifetime
MAX_CONTENT_LENGTH	No	200MB	Max request body size
KNOWLEDGE_MAX_UPLOAD_BYTES / KNOWLEDGE_UPLOAD_MAX_CHUNK_BYTES	No	200MB / 20MB	KB upload limits
FILES_MAX_UPLOAD_BYTES / FILES_UPLOAD_MAX_CHUNK_BYTES	No	100MB / 20MB	File manager upload limits
LOCAL_FINE_TUNE_API_URL	If fine-tuning	—	Upstream fine-tune service URL
LOCAL_FINE_TUNE_JWT_SECRET	If fine-tuning	change-me	HS256 secret for minted fine-tune JWTs
LOCAL_FINE_TUNING_CALL_ORIGIN	No	dev	origin claim sent to the fine-tune service
GDPR_DEFAULT_DELETE_MODE	No	soft	Default conversation deletion mode
CELERY_INCLUDE_ML_TASKS	No	true	Load ML libs / ML tasks in this worker (false for the prefork default worker)
CELERY_WORKER_CONCURRENCY	No	—	Worker concurrency override
CELERY_ENABLE_CHECK_SCHEDULED_WORKFLOW_RUNS_TASK	No	true	Enable scheduled-workflow tick
CELERY_ENABLE_RECONCILE_STUCK_WORKFLOW_RUNS_TASK	No	true	Enable stuck-run reconciliation
CELERY_ENABLE_CLEANUP_STALE_DIRECT_UPLOADS_TASK	No	true	Enable stale direct-upload cleanup
WORKFLOW_SCHEDULE_PENDING_MAX_AGE_SECONDS	No	900	Pending scheduled-run max age
WORKFLOW_SCHEDULE_RUNNING_MAX_AGE_SECONDS	No	7800	Running scheduled-run max age

Frontend (frontend/.env.template)

Change	Variable	Notes
Added	VITE_FILES_UPLOAD_SESSION_THRESHOLD_BYTES	Byte threshold to switch uploads to the chunked-session API (default 15MB)
Added	VITE_SSO_MICROSOFT_ENABLED	Show Microsoft SSO on login without calling the server status endpoint first
Removed	VITE_LOCAL_FINE_TUNE_API_URL	Fine-tune service URL moved server-side (LOCAL_FINE_TUNE_API_URL)

---

Infrastructure Changes

Celery worker topology (queue split)

The single Celery worker was split into two queues to make background processing both safe and scalable:

• default queue — I/O-bound tasks on the prefork pool (real concurrency). Safe because CELERY_INCLUDE_ML_TASKS=false keeps ML libraries (torch / sklearn / faiss) out of the master process, avoiding the OpenMP/MKL forked-mutex SIGSEGV.
• ml queue — ML pipeline / evaluation / local fine-tuning tasks on the solo pool (fork-unsafe libraries never fork); scale by adding replicas/pods.
• Routing is configured via task_routes; docker-compose.dev.yml runs split celery_worker_default (prefork, -Q default) and celery_worker_ml (solo, -Q ml) services. CI/CD and the owned build compose run a single solo worker consuming -Q default,ml.

Kubernetes

• New manifest backend/deploy/k8s/celery-workers.yaml — separate default (prefork) and ML (solo) worker Deployments for EKS.

Containers & health

• WebSocket service now pins image: genassist-ws-image in docker-compose.yml / docker-compose.dev.yml.
• Backend healthcheck relaxed: start_period 40s → 180s, retries 3 → 5 (accommodates longer startup, e.g. migrations).

---

Full Changelog: v1.7.1...v1.8.0

---

This discussion was created from the release v1.8.0.