Merge branch 'v3' of https://github.com/uzh-bf/klicker-uzh into v3-ai

Author: TheHummel

AGENTS.md

@@ -210,9 +210,11 @@ Without Traefik, use `http://localhost:<port>` directly. The `*.klicker.com` dom
 - **Infisical + Turbo env sync**: Any Infisical-managed env var used by tasks must be listed in `turbo.json` `globalEnv`; otherwise task runs/cache behavior can become stale or inconsistent across environments.
 - **Participant email uniqueness across auth modes**: Prisma enforces `Participant @@unique([email, isSSOAccount])`, so the same normalized email can exist once as manual and once as SSO. To block new cross-mode duplicates, account creation must explicitly check normalized email collisions in service logic. (`packages/graphql/src/services/accounts.ts`)
 - **Helm v3 secrets are external**: `deploy/charts/klicker-uzh-v3/` deployments reference `envFrom.secretRef` names, but the chart currently defines no `Secret` manifests; secrets must be provisioned out-of-band with matching names. (`deploy/charts/klicker-uzh-v3/templates/`)
+- **Production rollout strategy**: Do not use `Recreate` for production web/API Deployments; PDBs do not protect against Deployment-driven scale-downs, so slow image pulls can leave services with zero endpoints. Use `RollingUpdate` in `deploy/env-uzh-prd/values.yaml`, with `maxUnavailable: 0` only for singleton services and `1` for multi-replica services.
 - **Edited chat image hydration needs a stable source id**: Edited branch messages in `apps/chat` get fresh local message ids, so image hydration must distinguish the local target message id from the persisted source message id (`attachmentSourceMessageId`) when fetching and merging attachments. (`apps/chat/src/hooks/useThreadManagement.ts`, `apps/chat/src/stores/chatStore.ts`)
 - **Assistant UI chat drop targets**: `ComposerPrimitive.AttachmentDropzone` must wrap both normal and edit chat composer roots; it owns the drag/drop capture handlers that prevent native browser file navigation, while Klicker-specific limits stay in local composer code. (`apps/chat/src/components/thread.tsx`)
 - **Deployed chat model registries default attachment support off**: `apps/chat` loads `CHAT_MODEL_REGISTRY_JSON` via `chatModelRegistry.ts`, where omitted `supportsImageAttachments` values default to `false`; if deployment values override the built-in registry, each image-capable model must set the flag explicitly in `deploy/env-uzh-*/values.yaml` or the attach button disappears.
+- **OpenAI Responses storage and tool calls**: GPT-5.5 via the OpenAI Responses API can reference prior response items across tool-call steps; with `store: false`, LiteLLM/Azure can return "item not found" for those references. Keep `CHAT_OPENAI_STORE_RESPONSES=true` in shared/staged deployments that use Responses-compatible OpenAI backends, while local OpenRouter-style setups can leave it false. (`apps/chat/src/app/api/chatbots/[chatbotId]/chat/route.ts`, `deploy/env-uzh-*/values.yaml`)
 - **Embedded PWA messaging trust boundary**: For embedded PWA pages, use a parent-initiated `postMessage` handshake to capture `event.origin` and avoid `'*'` target origins; do not add a second per-platform messaging allowlist in page code. Embedding permission remains enforced separately by ingress `frame-ancestors`. (`apps/frontend-pwa/src/pages/course/[courseId]/practiceQuizzes/[id].tsx`, `deploy/charts/klicker-uzh-v3/templates/ingress-frontend-pwa.yaml`)
 - **Local embed harness target**: `util/embed-harness/` is for local verification only and should target the branch-local PWA (`http://127.0.0.1:3101/...`), not `https://pwa.klicker.com/...`, because production CSP / `frame-ancestors` blocks localhost embedding. (`util/embed-harness/`)
 - **Chat PWA login redirects**: `apps/chat/src/app/noLogin/page.tsx` must pass an absolute chat URL to the PWA login `redirect_to`; a relative chatbot path makes the PWA redirect to its own domain and 404. Local chat dev also needs ignored local env values for the backend `APP_SECRET` and `DATABASE_URL` so participant cookies verify and Prisma can load chatbot data.

CHANGELOG.md

@@ -2,6 +2,13 @@
 
 All notable changes to this project will be documented in this file. See [standard-version](https://github.com/conventional-changelog/standard-version) for commit guidelines.
 
+## [3.4.0-alpha.62](https://github.com/uzh-bf/klicker-uzh/compare/v3.4.0-alpha.61...v3.4.0-alpha.62) (2026-05-05)
+
+
+### Features
+
+* **chat:** make reasoning effort registry config-driven ([bb52d44](https://github.com/uzh-bf/klicker-uzh/commit/bb52d4449fe21fa828b38aa7410982e77d0eced0))
+
 ## [3.4.0-alpha.61](https://github.com/uzh-bf/klicker-uzh/compare/v3.4.0-alpha.60...v3.4.0-alpha.61) (2026-05-03)
 
 

apps/auth/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@klicker-uzh/auth",
-  "version": "3.4.0-alpha.61",
+  "version": "3.4.0-alpha.62",
   "license": "AGPL-3.0",
   "dependencies": {
     "@auth/prisma-adapter": "2.10.0",

apps/backend-docker/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@klicker-uzh/backend-docker",
-  "version": "3.4.0-alpha.61",
+  "version": "3.4.0-alpha.62",
   "license": "AGPL-3.0",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",

apps/chat/.env.local.example

@@ -2,6 +2,7 @@
 AZURE_RESOURCE_NAME=klicker-ai
 AZURE_API_KEY=sk-XXX
 AZURE_RESPONSES_API_VERSION=preview
+CHAT_OPENAI_STORE_RESPONSES=false
 
 # MCP Key
 MCP_KEY=sk-XXX

apps/chat/src/app/api/chatbots/[chatbotId]/chat/route.ts

@@ -5,6 +5,7 @@ import {
   getChatModelRegistry,
 } from '@/src/lib/server/chatModelRegistry'
 import { ensureImagePreviewBase64 } from '@/src/lib/server/imagePreview'
+import { getOpenAIResponsesStore } from '@/src/lib/server/openaiResponsesOptions'
 import {
   getAggregatedMCPTools,
   type MCPServerWithConfig,
@@ -24,10 +25,7 @@ import {
 import { createHash, randomUUID } from 'crypto'
 import { NextRequest, NextResponse } from 'next/server'
 import { DEFAULT_PROMPT } from 'src/lib/config/prompts'
-import {
-  REASONING_EFFORT_OPTIONS,
-  type ReasoningEffort,
-} from 'src/lib/config/reasoning'
+import { type ReasoningEffort } from 'src/lib/config/reasoning'
 import { CreditsService } from 'src/services/credits'
 import { DisclaimersService } from 'src/services/disclaimers'
 import {
@@ -687,10 +685,7 @@ export async function POST(
       .optional()
       .transform((val) => val?.toLowerCase())
       .default('tutor'),
-    reasoningEffort: z
-      .enum(REASONING_EFFORT_OPTIONS)
-      .optional()
-      .default('none'),
+    reasoningEffort: z.string().min(1).optional().default('none'),
     parentId: z.string().min(1).nullable().optional(),
     assistantMessageId: z.string().min(1),
     images: z
@@ -1359,7 +1354,7 @@ export async function POST(
     experimental_telemetry: { isEnabled: true },
     providerOptions: {
       openai: {
-        store: false,
+        store: getOpenAIResponsesStore(),
         ...(providerReasoningEffort && {
           reasoningEffort: providerReasoningEffort,
           reasoningSummary: 'auto',

apps/chat/src/hooks/useChatResponse.ts

@@ -2,10 +2,7 @@ import { useParams } from 'next/navigation'
 import { useCallback, useRef } from 'react'
 import { hasAllImageAttachmentsHydrated } from '../lib/attachments/attachmentState'
 import { authedFetch } from '../lib/client/authedFetch'
-import {
-  REASONING_EFFORT_OPTIONS,
-  type ReasoningEffort,
-} from '../lib/config/reasoning'
+import { type ReasoningEffort } from '../lib/config/reasoning'
 import { generateId } from '../lib/utils/chatUtils'
 import {
   useChatStore,
@@ -474,10 +471,8 @@ export function useChatResponse(
                     >
                     const reasoningEffort =
                       typeof metadata.reasoningEffort === 'string' &&
-                      REASONING_EFFORT_OPTIONS.includes(
-                        metadata.reasoningEffort as ReasoningEffort
-                      )
-                        ? (metadata.reasoningEffort as ReasoningEffort)
+                      metadata.reasoningEffort.length > 0
+                        ? metadata.reasoningEffort
                         : null
 
                     messageMetadata = {

apps/chat/src/lib/config/reasoning.ts

@@ -1,10 +1 @@
-export const REASONING_EFFORT_OPTIONS = [
-  'none',
-  'minimal',
-  'low',
-  'medium',
-  'high',
-  'xhigh',
-] as const
-
-export type ReasoningEffort = (typeof REASONING_EFFORT_OPTIONS)[number]
+export type ReasoningEffort = string

apps/chat/src/lib/server/chatModelRegistry.ts

@@ -1,27 +1,36 @@
 import { z } from 'zod'
-import {
-  REASONING_EFFORT_OPTIONS,
-  type ReasoningEffort,
-} from '../config/reasoning'
-
-const reasoningEffortSchema = z.enum(REASONING_EFFORT_OPTIONS)
-
-const chatModelSchema = z.object({
-  id: z.string().min(1),
-  deploymentId: z.string().min(1),
-  name: z.string().min(1),
-  description: z.string().default(''),
-  fallback: z.boolean().default(false),
-  supportsReasoning: z.boolean().default(false),
-  supportsImageAttachments: z.boolean().default(false),
-  supportedReasoningEfforts: z.array(reasoningEffortSchema).optional(),
-  maxOutputTokens: z.number().positive().optional(),
-  apiVersion: z.string().min(1).optional(),
-  cost: z.object({
-    input: z.number().nonnegative(),
-    output: z.number().nonnegative(),
-  }),
-})
+import { type ReasoningEffort } from '../config/reasoning'
+
+const chatModelSchema = z
+  .object({
+    id: z.string().min(1),
+    deploymentId: z.string().min(1),
+    name: z.string().min(1),
+    description: z.string().default(''),
+    fallback: z.boolean().default(false),
+    supportsReasoning: z.boolean().default(false),
+    supportsImageAttachments: z.boolean().default(false),
+    supportedReasoningEfforts: z.array(z.string().min(1)).optional(),
+    maxOutputTokens: z.number().positive().optional(),
+    apiVersion: z.string().min(1).optional(),
+    cost: z.object({
+      input: z.number().nonnegative(),
+      output: z.number().nonnegative(),
+    }),
+  })
+  .superRefine((model, ctx) => {
+    if (
+      model.supportsReasoning &&
+      (!model.supportedReasoningEfforts ||
+        model.supportedReasoningEfforts.length === 0)
+    ) {
+      ctx.addIssue({
+        code: z.ZodIssueCode.custom,
+        path: ['supportedReasoningEfforts'],
+        message: `Model "${model.id}" has supportsReasoning=true but no supportedReasoningEfforts configured.`,
+      })
+    }
+  })
 
 const chatModelRegistrySchema = z
   .array(chatModelSchema)
@@ -63,55 +72,8 @@ export type ChatModelConfig = Omit<
   supportedReasoningEfforts: ReasoningEffort[]
 }
 
-const BASE_REASONING_EFFORTS: ReasoningEffort[] = [
-  'none',
-  'minimal',
-  'low',
-  'medium',
-  'high',
-]
-const FRONTIER_REASONING_EFFORTS: ReasoningEffort[] = [
-  'none',
-  'low',
-  'medium',
-  'high',
-  'xhigh',
-]
-const GPT5_REASONING_EFFORTS: ReasoningEffort[] = [
-  'minimal',
-  'low',
-  'medium',
-  'high',
-]
-
-function dedupeReasoningEfforts(efforts: readonly ReasoningEffort[]) {
-  const seen = new Set<ReasoningEffort>()
-  const deduped: ReasoningEffort[] = []
-  for (const effort of efforts) {
-    if (seen.has(effort)) continue
-    seen.add(effort)
-    deduped.push(effort)
-  }
-  return deduped
-}
-
-function getDefaultReasoningEffortsForModel(
-  modelId: string
-): ReasoningEffort[] {
-  const normalizedId = modelId.toLowerCase()
-  if (
-    normalizedId.startsWith('gpt-5.5') ||
-    normalizedId.startsWith('gpt-5.4')
-  ) {
-    return [...FRONTIER_REASONING_EFFORTS]
-  }
-  if (normalizedId.startsWith('gpt-5.1')) {
-    return [...BASE_REASONING_EFFORTS]
-  }
-  if (normalizedId.startsWith('gpt-5')) {
-    return [...GPT5_REASONING_EFFORTS]
-  }
-  return [...BASE_REASONING_EFFORTS]
+function dedupeStrings(values: readonly string[]) {
+  return Array.from(new Set(values))
 }
 
 function normalizeChatModelConfig(model: RawChatModelConfig): ChatModelConfig {
@@ -122,15 +84,11 @@ function normalizeChatModelConfig(model: RawChatModelConfig): ChatModelConfig {
     }
   }
 
-  const providedEfforts = model.supportedReasoningEfforts ?? []
-  const normalizedEfforts =
-    providedEfforts.length > 0
-      ? dedupeReasoningEfforts(providedEfforts)
-      : getDefaultReasoningEffortsForModel(model.id)
-
   return {
     ...model,
-    supportedReasoningEfforts: normalizedEfforts,
+    supportedReasoningEfforts: dedupeStrings(
+      model.supportedReasoningEfforts ?? []
+    ),
   }
 }
 
@@ -149,7 +107,7 @@ const DEFAULT_MODEL_REGISTRY: ChatModelConfig[] = [
     fallback: false,
     supportsReasoning: true,
     supportsImageAttachments: true,
-    supportedReasoningEfforts: [...FRONTIER_REASONING_EFFORTS],
+    supportedReasoningEfforts: ['none', 'low', 'medium', 'high', 'xhigh'],
     maxOutputTokens: 2048,
     cost: { input: 5.0, output: 30.0 },
   },
@@ -161,7 +119,7 @@ const DEFAULT_MODEL_REGISTRY: ChatModelConfig[] = [
     fallback: false,
     supportsReasoning: true,
     supportsImageAttachments: true,
-    supportedReasoningEfforts: [...FRONTIER_REASONING_EFFORTS],
+    supportedReasoningEfforts: ['none', 'low', 'medium', 'high', 'xhigh'],
     maxOutputTokens: 2048,
     cost: { input: 2.5, output: 15.0 },
   },
@@ -173,7 +131,7 @@ const DEFAULT_MODEL_REGISTRY: ChatModelConfig[] = [
     fallback: false,
     supportsReasoning: true,
     supportsImageAttachments: true,
-    supportedReasoningEfforts: [...BASE_REASONING_EFFORTS],
+    supportedReasoningEfforts: ['none', 'minimal', 'low', 'medium', 'high'],
     maxOutputTokens: 2048,
     cost: { input: 1.25, output: 10.0 },
   },
@@ -239,11 +197,10 @@ export function parseReasoningEffortByModel(
     if (!Array.isArray(rawEfforts)) continue
     const validEfforts: ReasoningEffort[] = rawEfforts.filter(
       (effort): effort is ReasoningEffort =>
-        typeof effort === 'string' &&
-        REASONING_EFFORT_OPTIONS.includes(effort as ReasoningEffort)
+        typeof effort === 'string' && effort.length > 0
     )
 
-    const dedupedEfforts = dedupeReasoningEfforts(validEfforts)
+    const dedupedEfforts = dedupeStrings(validEfforts)
     if (dedupedEfforts.length === 0) continue
 
     result[modelId] = dedupedEfforts
@@ -261,9 +218,7 @@ export function getAllowedReasoningEffortsForModel(
 ): ReasoningEffort[] {
   if (!model.supportsReasoning) return []
 
-  const supportedEfforts = dedupeReasoningEfforts(
-    model.supportedReasoningEfforts
-  )
+  const supportedEfforts = dedupeStrings(model.supportedReasoningEfforts)
   if (supportedEfforts.length === 0) return []
 
   const configuredByModel = parseReasoningEffortByModel(rawConfig)

apps/chat/src/lib/server/openaiResponsesOptions.ts

@@ -0,0 +1,10 @@
+const TRUTHY_VALUES = new Set(['true', '1', 'yes', 'on'])
+
+export function parseOpenAIResponsesStore(value: string | undefined): boolean {
+  if (!value) return false
+  return TRUTHY_VALUES.has(value.trim().toLowerCase())
+}
+
+export function getOpenAIResponsesStore(): boolean {
+  return parseOpenAIResponsesStore(process.env.CHAT_OPENAI_STORE_RESPONSES)
+}