API: jobbstatus-endpoint for polling av AI-behandling

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

CLAUDE.md

@@ -5,6 +5,9 @@ Sidelinja er et redaksjonelt operativsystem og kunnskapsgraf for podcast-produks
 Self-hosted på Hetzner VPS med full datakontroll.
 
 ## Arbeidsflyt
+- **Standard arbeidsmodus:** Start i planleggingsmodus. Lag en grundig plan, få godkjenning, deretter implementer. Jobbene er ment å kunne kjøre lenge autonomt uten input underveis.
+- **Testmiljø:** `./dev.sh` er den kanoniske måten å starte utviklingsmiljøet. Når nye tjenester, steg eller oppsett-quirks oppdages, oppdater alltid `dev.sh` slik at kunnskapen bevares i scriptet — ikke bare i hodet. Før Vegard tester i browser: kjør `./dev.sh`, verifiser med `cargo check`/`svelte-check`/`curl`, og meld tilbake at det er klart.
+- **Browser-testing:** Claude har ikke tilgang til browser. Visuell testing og interaksjon gjøres av Vegard. Claude kan verifisere backend (kompilering, API-kall, database-state) men ikke frontend-rendering.
 - **Commit og push:** Bruk egen vurdering. Commit når arbeidet er logisk komplett, push til Forgejo når det gir mening. Ingen grunn til å spørre — det er trygt og reverserbart.
 - **Deploy til produksjon:** Krever alltid eksplisitt godkjenning fra Vegard. Deploy = SSH til server + pull + docker compose up. Aldri gjør dette uten å spørre først.
 - **Diskusjon:** Forklar og diskuter før arkitekturendringer eller uvanlige valg. For implementering innenfor eksisterende spec — bare kjør.

docs/erfaringer/spacetimedb_integrasjon.md

@@ -107,5 +107,26 @@ Ny modell:
 - SyncOutbox-events prosesseres hver 1. sekund
 - Støtter: `messages/insert`, `messages/delete`, `messages/update`, `message_reactions/insert`, `message_reactions/delete`
 
+## 7. Subscription-begrensninger
+
+**SpacetimeDB-subscriptions støtter IKKE JOINs.** En subscription-query som `SELECT mr.* FROM message_reaction mr JOIN chat_message cm ON cm.id = mr.message_id WHERE ...` feiler stille — `onApplied` kalles aldri, og ingen data vises.
+
+Bruk kun enkle `SELECT * FROM tabell WHERE ...`-queries i `.subscribe([...])`. Filtrer heller klient-side etter at data er lastet.
+
+Eksempel:
+```typescript
+// FEIL — feiler stille, ingen data
+.subscribe([
+  `SELECT * FROM chat_message WHERE channel_id = '${id}'`,
+  `SELECT mr.* FROM message_reaction mr JOIN chat_message cm ON cm.id = mr.message_id WHERE cm.channel_id = '${id}'`
+]);
+
+// RIKTIG — last alle reaksjoner, filtrer i koden
+.subscribe([
+  `SELECT * FROM chat_message WHERE channel_id = '${id}'`,
+  `SELECT * FROM message_reaction`
+]);
+```
+
 ### Fallback
 PG-polling adapter (`pg.svelte.ts`) brukes kun når SpacetimeDB ikke er konfigurert. Markeres som `readonly: true`.

docs/features/chat.md

@@ -86,13 +86,16 @@ Kun aktive når `config.mentions = true`.
 * **Mobil-optimalisert:** Autocomplete-listen er tappbar og tilpasset mindre skjermer.
 
 ## 5. Tråder
-Kun aktive når `config.threads = true`. Meldinger kan ha en `reply_to`-referanse. Frontend viser tråder som innrykk eller ekspanderbare grupper.
+Kun aktive når `config.threads = true`. Meldinger kan ha en `reply_to`-referanse. Frontend grupperer meldinger i tråder (rot + svar) med visuell skillelinje mellom hver tråd. Svar vises med innrykk og vertikal linje under rot-meldingen, uten ekstra skillelinje mellom rot og svar.
 
 ## 6. Vedlegg
 Kun aktive når `config.attachments = true`. Meldinger kan ha vedlegg via `message_attachments` → `media_files`. Whiteboard-eksport kan knyttes som vedlegg.
 
 ## 7. Versjonshistorikk
-Alle meldinger støtter redigering med full historikk via `message_revisions`. Original tekst bevares alltid.
+Alle meldinger støtter redigering med full historikk via `message_revisions`. Original tekst bevares alltid. AI-behandlede meldinger har en revisjons-toggle i UI — brukeren kan veksle mellom AI-versjon og original tekst. AI-output rendres som Markdown via `marked`.
+
+## 7.1 Meldingsvisning
+Lange meldinger (mer enn 2 linjer) kollapses automatisk med en "Vis mer"-knapp. Ved ekspandering vises "Vis mindre" både over og under meldingen, slik at man slipper å scrolle for å kollapse igjen.
 
 ## 8. Tale-til-tekst (Voice-to-text)
 Mobilvennlig diktering for situasjoner der tastatur er upraktisk. Brukeren trykker en mikrofon-knapp, snakker, og får teksten tilbake som en vanlig melding klar til redigering og sending.
@@ -124,8 +127,11 @@ Channels med `config.ttl_days` satt til et tall får sine meldinger automatisk s
 - **Worker warmup (`worker/src/warmup.rs`):** PG → SpacetimeDB ved oppstart. Per-kanal konfig (all/messages/days/none). Trådbasert henting.
 - **Worker sync (`worker/src/sync.rs`):** SpacetimeDB → PG hvert sekund. Insert/delete/update meldinger + reaksjoner.
 - **Admin-side (`/admin/channels`):** Per-kanal warmup-konfigurasjon.
-- **Tråder:** Komplett trådvisning med datogruppering og autoscroll.
+- **Tråder:** Komplett trådvisning med datogruppering, autoscroll og visuell skillelinje mellom tråder.
 - **Reaksjoner:** Via SpacetimeDB-reducers, synket til PG.
+- **Meldingskollaps:** Lange meldinger begrenses til 2 linjer med "Vis mer"/"Vis mindre".
+- **AI-behandling:** Meldinger kan AI-behandles (✨-knapp). Revisjons-toggle viser original vs. AI-versjon. Markdown-rendering for AI-output.
+- **Konvertering:** Meldinger kan opprettes som kanban-kort eller kalenderhendelse (dialog sier "Opprett", ikke "Konverter" — meldingen beholdes i chatten).
 
 ### Gjenstår
 - **Vedlegg, TTL** — avventer implementering.

docs/infra/ai_gateway.md

@@ -24,59 +24,86 @@ Fordeler:
 
 ## 3. Modellruting
 
-Modellvalg styres av to mekanismer:
+### 3.1 Arkitekturprinsipp: PG eier config, LiteLLM er stateløs
 
-### 3.1 Standard ruting (config.yaml)
+PostgreSQL er single source of truth for all modellkonfigurasjon. LiteLLM er en stateløs proxy som får generert `config.yaml` fra PG-data. Dette gir:
-LiteLLM konfigureres med modellaliaser som mapper til billigste egnede leverandør:
 
-```yaml
+* **Ingen avhengighet til LiteLLM sitt admin API** — de endrer API mellom versjoner
-model_list:
+* **All konfig i samme backup/migrasjon** som resten av systemet
-  # Ruting: billigste først, fallback til dyrere
+* **Enkel bytte** — hvis LiteLLM erstattes, er all konfig intakt i PG
-  - model_name: "sidelinja/rutine"
+* **Admin-UI i SvelteKit** — gjenbruker eksisterende `/admin/`-mønster
-    litellm_params:
-      model: "gemini/gemini-2.0-flash"
-      api_key: "os.environ/GEMINI_API_KEY"
-  - model_name: "sidelinja/rutine"
-    litellm_params:
-      model: "openrouter/google/gemini-2.0-flash-001"
-      api_key: "os.environ/OPENROUTER_API_KEY"
 
-  - model_name: "sidelinja/resonering"
+### 3.2 Datamodell
-    litellm_params:
-      model: "anthropic/claude-sonnet-4-20250514"
-      api_key: "os.environ/ANTHROPIC_API_KEY"
-  - model_name: "sidelinja/resonering"
-    litellm_params:
-      model: "openrouter/anthropic/claude-sonnet-4-20250514"
-      api_key: "os.environ/OPENROUTER_API_KEY"
 
-router_settings:
+```sql
-  routing_strategy: "simple-shuffle"  # prøv første, fallback til neste
+-- Globale modellaliaser (server-nivå, ikke per workspace)
-  num_retries: 2
+CREATE TABLE ai_model_aliases (
-  timeout: 60
+    id            UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+    alias         TEXT NOT NULL,           -- 'sidelinja/rutine', 'sidelinja/resonering'
+    description   TEXT,                    -- 'Billig, høyt volum'
+    is_active     BOOLEAN NOT NULL DEFAULT true,
+    created_at    TIMESTAMPTZ NOT NULL DEFAULT now(),
+    UNIQUE(alias)
+);
 
-general_settings:
+-- Leverandør-modeller med prioritert fallback per alias
-  master_key: "os.environ/LITELLM_MASTER_KEY"
+CREATE TABLE ai_model_providers (
+    id            UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+    alias_id      UUID NOT NULL REFERENCES ai_model_aliases(id) ON DELETE CASCADE,
+    provider      TEXT NOT NULL,           -- 'gemini', 'openrouter', 'anthropic'
+    model         TEXT NOT NULL,           -- 'gemini/gemini-2.5-flash', 'openrouter/anthropic/claude-sonnet-4'
+    api_key_env   TEXT NOT NULL,           -- 'GEMINI_API_KEY', 'OPENROUTER_API_KEY'
+    priority      SMALLINT NOT NULL,       -- lavere = prøves først
+    is_active     BOOLEAN NOT NULL DEFAULT true,
+    UNIQUE(alias_id, model)
+);
+
+-- Jobbtype → modellalias mapping
+CREATE TABLE ai_job_routing (
+    job_type      TEXT PRIMARY KEY,        -- 'ai_text_process', 'whisper_postprocess', etc.
+    alias         TEXT NOT NULL,           -- 'sidelinja/rutine'
+    description   TEXT
+);
 ```
 
-### 3.2 Jobbkø-styrt modellvalg
+### 3.3 Config-generering
-Jobbkøen (se `jobbkø.md`) spesifiserer modellalias per jobbtype:
 
-| Jobbtype | Modellalias | Begrunnelse |
+SvelteKit-serveren genererer `config.yaml` fra PG ved oppstart og ved endringer i admin-panelet:
+
+1. Les aktive aliaser og deres providers (sortert etter priority)
+2. Skriv `config.yaml` til volum delt med LiteLLM-containeren
+3. Restart LiteLLM (`docker restart ai-gateway`) eller send `SIGHUP`
+
+Generert config inkluderer alltid `router_settings` og `general_settings` fra faste verdier — kun `model_list` er dynamisk.
+
+### 3.4 Jobbkø-styrt modellvalg
+
+Jobbkøen bruker `ai_job_routing` for å bestemme modellalias per jobbtype:
+
+| Jobbtype | Standard alias | Begrunnelse |
 |---|---|---|
-| `whisper_postprocess` (transkripsjonsvasking) | `sidelinja/rutine` | Høyt volum, lav kompleksitet |
+| `ai_text_process` (✨-behandling) | `sidelinja/rutine` | Tekstvasking, høyt volum |
-| `openrouter_analyze` (metadata-uttrekk) | `sidelinja/rutine` | Strukturert output, lav kompleksitet |
+| `whisper_postprocess` | `sidelinja/rutine` | Transkripsjonsvasking, høyt volum |
-| `research_clip` (research-oppsummering) | `sidelinja/rutine` | Høyt volum |
+| `research_clip` | `sidelinja/rutine` | Research-oppsummering, høyt volum |
-| `live_factoid_eval` (live-assistent) | `sidelinja/resonering` | Krever presis vurdering under tidspress |
+| `live_factoid_eval` | `sidelinja/resonering` | Krever presis vurdering under tidspress |
 
 Modellalias lagres som felt på jobben i PG — kan overstyres manuelt per jobb ved behov.
 
+### 3.5 Admin-panel (`/admin/ai`)
+
+Admin-panelet lar administrator:
+* Se og redigere modellaliaser og deres fallback-liste (drag-and-drop prioritering)
+* Aktivere/deaktivere individuelle leverandør-modeller
+* Endre jobbtype → alias mapping
+* Se live-status: hvilke leverandører som svarer, responstider
+* Trigge config-regenerering og LiteLLM-restart
+
 ## 4. Docker-oppsett
 
 ```yaml
 # docker-compose.dev.yml / docker-compose.yml
 ai-gateway:
-  image: ghcr.io/berriai/litellm:main
+  image: ghcr.io/berriai/litellm:main-stable
   restart: unless-stopped
   command: --config /etc/litellm/config.yaml
   environment:
@@ -151,36 +178,55 @@ tests/prompts/
     └── dataset.json
 ```
 
-## 6. Kostnadskontroll
+## 6. Tokenregnskap og kostnadskontroll
 
-LiteLLM har innebygd logging, men mangler workspace-nivå budsjettering. For å forhindre kostnadssprekk:
+### 6.1 Token-logging per workspace
 
-### 6.1 Workspace-budsjett
+Rust-workeren logger tokenforbruk etter hvert AI-kall. Dataen lagres i PG:
-Hver workspace har et månedlig AI-budsjett lagret i `workspaces.settings` (JSONB):
 
-```json
+```sql
-{
+CREATE TABLE ai_usage_log (
-  "ai_budget": {
+    id              UUID PRIMARY KEY DEFAULT gen_random_uuid(),
-    "monthly_limit_usd": 50,
+    workspace_id    UUID NOT NULL REFERENCES workspaces(id) ON DELETE CASCADE,
-    "alert_threshold_pct": 80,
+    job_id          UUID REFERENCES job_queue(id) ON DELETE SET NULL,
-    "auto_fallback": true
+    model_alias     TEXT NOT NULL,          -- 'sidelinja/rutine'
-  }
+    model_actual    TEXT,                   -- 'gemini/gemini-2.5-flash' (fra LiteLLM-respons)
-}
+    prompt_tokens   INT NOT NULL,
+    completion_tokens INT NOT NULL,
+    total_tokens    INT NOT NULL,
+    estimated_cost  NUMERIC(10, 6),         -- USD, beregnet fra kjente priser
+    job_type        TEXT,                   -- 'ai_text_process', etc.
+    created_at      TIMESTAMPTZ NOT NULL DEFAULT now()
+);
+
+CREATE INDEX idx_ai_usage_workspace_month ON ai_usage_log (workspace_id, created_at);
 ```
 
-- **Sporing:** SvelteKit logger token-bruk per AI-kall med workspace_id og jobbtype i `ai_usage_log`-tabellen (flyktig, TTL 90 dager).
+**Flyten:**
-- **Alert:** Når 80 % av budsjettet er brukt, postes varsel i workspace-chat (system-channel).
+1. Rust-worker sender AI-kall via gateway, får tilbake `usage` i responsen
-- **Auto-fallback:** Når budsjettet er nådd og `auto_fallback: true`, rutes alle kall til `sidelinja/rutine` (billigste modell). Ellers blokkeres AI-kall med feilmelding.
+2. Worker skriver rad til `ai_usage_log` med workspace_id, tokens og modellinfo
+3. Estimert kostnad beregnes fra en enkel prisliste i config (oppdateres manuelt)
 
-### 6.2 Per-episode maks-kostnad
+### 6.2 Visning — to nivåer
+
+**Admin (`/admin/ai`):**
+Aggregert oversikt over alle workspaces. Tabell med totaler per workspace/modell/periode. Identifiserer kostnadsdrivere.
+
+**Workspace (sidebar-widget):**
+Enkel tekst-indikator i workspace-sidebar: `✨ 12.4k tokens denne uken`. Klikk åpner detaljert visning med fordeling per jobbtype og modell. Ingen speedometer — det krever et definert budsjett for å gi mening, og det er overkill for MVP.
+
+### 6.3 Workspace-budsjett (fase 2)
+
+Når token-logging er på plass, kan budsjett-tak legges til:
+
+- Budsjett lagres i `workspaces.settings` (JSONB): `{ "ai_budget": { "monthly_limit_usd": 50 } }`
+- Rust-worker sjekker aggregert forbruk før AI-kall
+- Ved budsjett nær: fall tilbake til `sidelinja/rutine` (billigste)
+- Ved budsjett nådd: sett jobb i `paused` med varsel i workspace-chat
+
+### 6.4 Per-episode maks-kostnad
 Podcastfabrikken-jobber (whisper + metadata + oppsummering) kan estimere totalkostnad basert på lydlengde. Jobben avbrytes med varsel hvis estimert kostnad overstiger `max_cost_per_episode` (default: $5).
 
-### 6.3 Modell-nedgradering
-Jobbkøen støtter automatisk modell-nedgradering ved kostnadsmål:
-1. Prøv `sidelinja/resonering` (Claude)
-2. Ved budsjett-nær: fall tilbake til `sidelinja/rutine` (Gemini gratis)
-3. Ved budsjett-nådd: sett jobb i `paused`-status med varsel
-
 ## 7. Dataklassifisering (ref. docs/arkitektur.md 2.2)
 
 | Data | Kategori | Detaljer |

migrations/0015_usage_action_column.sql

@@ -0,0 +1,8 @@
+-- 0015_usage_action_column.sql
+-- Legg til action-kolonne i ai_usage_log for å spore hvilken prompt som ble brukt.
+
+BEGIN;
+
+ALTER TABLE ai_usage_log ADD COLUMN IF NOT EXISTS action TEXT;
+
+COMMIT;

web/src/lib/blocks/KanbanBlock.svelte

@@ -4,6 +4,7 @@
 	import type { KanbanConnection, KanbanColumn } from '$lib/kanban/types';
 	import type { MessageData } from '$lib/types/message';
 	import MessageBox from '$lib/components/MessageBox.svelte';
+	import Editor from '$lib/components/Editor.svelte';
 
 	let { props = {} }: { props?: Record<string, unknown> } = $props();
 
@@ -225,12 +226,13 @@
 					bind:value={editTitle}
 					placeholder="Tittel"
 				/>
-				<textarea
+				<div class="edit-desc-editor">
-					class="edit-desc"
+					<Editor
-					bind:value={editDescription}
+						mode="extended"
-					placeholder="Beskrivelse (valgfritt)"
+						bind:content={editDescription}
-					rows="3"
+						placeholder="Beskrivelse (valgfritt)"
-				></textarea>
+					/>
+				</div>
 				<div class="modal-actions">
 					<button type="button" onclick={saveEdit}>Lagre</button>
 					<button type="button" class="delete" onclick={handleDeleteCard}>Slett</button>
@@ -457,7 +459,7 @@
 		border: 1px solid #2d3148;
 		border-radius: 10px;
 		padding: 1rem;
-		width: 340px;
+		width: 420px;
 		max-width: 90vw;
 		display: flex;
 		flex-direction: column;
@@ -480,24 +482,13 @@
 		border-color: #3b82f6;
 	}
 
-	.edit-desc {
+	.edit-desc-editor {
-		background: #0f1117;
 		border: 1px solid #2d3148;
 		border-radius: 6px;
-		color: #e1e4e8;
+		overflow: hidden;
-		padding: 0.5rem;
-		font-size: 0.8rem;
-		font-family: inherit;
-		resize: vertical;
-		line-height: 1.4;
 	}
 
-	.edit-desc:focus {
+	.edit-title::placeholder {
-		outline: none;
-		border-color: #3b82f6;
-	}
-
-	.edit-desc::placeholder, .edit-title::placeholder {
 		color: #8b92a5;
 	}
 

web/src/lib/chat/module_bindings/add_reaction_reducer.ts

@@ -1,19 +1,19 @@
-// THIS FILE IS AUTOMATICALLY GENERATED BY SPACETIMEDB. EDITS TO THIS FILE
+// THIS FILE IS AUTOMATICALLY GENERATED BY SPACETIMEDB. EDITS TO THIS FILE
-// WILL NOT BE SAVED. MODIFY TABLES IN YOUR MODULE SOURCE CODE INSTEAD.
+// WILL NOT BE SAVED. MODIFY TABLES IN YOUR MODULE SOURCE CODE INSTEAD.
-
+
-/* eslint-disable */
+/* eslint-disable */
-/* tslint:disable */
+/* tslint:disable */
-import {
+import {
-  TypeBuilder as __TypeBuilder,
+  TypeBuilder as __TypeBuilder,
-  t as __t,
+  t as __t,
-  type AlgebraicTypeType as __AlgebraicTypeType,
+  type AlgebraicTypeType as __AlgebraicTypeType,
-  type Infer as __Infer,
+  type Infer as __Infer,
-} from "spacetimedb";
+} from "spacetimedb";
-
+
-export default {
+export default {
-  messageId: __t.string(),
+  messageId: __t.string(),
-  workspaceId: __t.string(),
+  workspaceId: __t.string(),
-  userId: __t.string(),
+  userId: __t.string(),
-  userName: __t.string(),
+  userName: __t.string(),
-  reaction: __t.string(),
+  reaction: __t.string(),
-};
+};

web/src/lib/chat/module_bindings/clear_channel_reducer.ts

@@ -1,15 +1,15 @@
-// THIS FILE IS AUTOMATICALLY GENERATED BY SPACETIMEDB. EDITS TO THIS FILE
+// THIS FILE IS AUTOMATICALLY GENERATED BY SPACETIMEDB. EDITS TO THIS FILE
-// WILL NOT BE SAVED. MODIFY TABLES IN YOUR MODULE SOURCE CODE INSTEAD.
+// WILL NOT BE SAVED. MODIFY TABLES IN YOUR MODULE SOURCE CODE INSTEAD.
-
+
-/* eslint-disable */
+/* eslint-disable */
-/* tslint:disable */
+/* tslint:disable */
-import {
+import {
-  TypeBuilder as __TypeBuilder,
+  TypeBuilder as __TypeBuilder,
-  t as __t,
+  t as __t,
-  type AlgebraicTypeType as __AlgebraicTypeType,
+  type AlgebraicTypeType as __AlgebraicTypeType,
-  type Infer as __Infer,
+  type Infer as __Infer,
-} from "spacetimedb";
+} from "spacetimedb";
-
+
-export default {
+export default {
-  channelId: __t.string(),
+  channelId: __t.string(),
-};
+};

web/src/lib/chat/module_bindings/edit_message_reducer.ts

@@ -1,17 +1,17 @@
-// THIS FILE IS AUTOMATICALLY GENERATED BY SPACETIMEDB. EDITS TO THIS FILE
+// THIS FILE IS AUTOMATICALLY GENERATED BY SPACETIMEDB. EDITS TO THIS FILE
-// WILL NOT BE SAVED. MODIFY TABLES IN YOUR MODULE SOURCE CODE INSTEAD.
+// WILL NOT BE SAVED. MODIFY TABLES IN YOUR MODULE SOURCE CODE INSTEAD.
-
+
-/* eslint-disable */
+/* eslint-disable */
-/* tslint:disable */
+/* tslint:disable */
-import {
+import {
-  TypeBuilder as __TypeBuilder,
+  TypeBuilder as __TypeBuilder,
-  t as __t,
+  t as __t,
-  type AlgebraicTypeType as __AlgebraicTypeType,
+  type AlgebraicTypeType as __AlgebraicTypeType,
-  type Infer as __Infer,
+  type Infer as __Infer,
-} from "spacetimedb";
+} from "spacetimedb";
-
+
-export default {
+export default {
-  id: __t.string(),
+  id: __t.string(),
-  workspaceId: __t.string(),
+  workspaceId: __t.string(),
-  newBody: __t.string(),
+  newBody: __t.string(),
-};
+};

web/src/lib/chat/module_bindings/index.ts

@@ -156,3 +156,4 @@ export class DbConnection extends __DbConnectionImpl<typeof REMOTE_MODULE> {
     return new SubscriptionBuilder(this);
   };
 }
+

web/src/lib/chat/module_bindings/load_reactions_reducer.ts

@@ -1,15 +1,15 @@
-// THIS FILE IS AUTOMATICALLY GENERATED BY SPACETIMEDB. EDITS TO THIS FILE
+// THIS FILE IS AUTOMATICALLY GENERATED BY SPACETIMEDB. EDITS TO THIS FILE
-// WILL NOT BE SAVED. MODIFY TABLES IN YOUR MODULE SOURCE CODE INSTEAD.
+// WILL NOT BE SAVED. MODIFY TABLES IN YOUR MODULE SOURCE CODE INSTEAD.
-
+
-/* eslint-disable */
+/* eslint-disable */
-/* tslint:disable */
+/* tslint:disable */
-import {
+import {
-  TypeBuilder as __TypeBuilder,
+  TypeBuilder as __TypeBuilder,
-  t as __t,
+  t as __t,
-  type AlgebraicTypeType as __AlgebraicTypeType,
+  type AlgebraicTypeType as __AlgebraicTypeType,
-  type Infer as __Infer,
+  type Infer as __Infer,
-} from "spacetimedb";
+} from "spacetimedb";
-
+
-export default {
+export default {
-  reactionsJson: __t.string(),
+  reactionsJson: __t.string(),
-};
+};

web/src/lib/chat/module_bindings/message_reaction_table.ts

@@ -1,19 +1,19 @@
-// THIS FILE IS AUTOMATICALLY GENERATED BY SPACETIMEDB. EDITS TO THIS FILE
+// THIS FILE IS AUTOMATICALLY GENERATED BY SPACETIMEDB. EDITS TO THIS FILE
-// WILL NOT BE SAVED. MODIFY TABLES IN YOUR MODULE SOURCE CODE INSTEAD.
+// WILL NOT BE SAVED. MODIFY TABLES IN YOUR MODULE SOURCE CODE INSTEAD.
-
+
-/* eslint-disable */
+/* eslint-disable */
-/* tslint:disable */
+/* tslint:disable */
-import {
+import {
-  TypeBuilder as __TypeBuilder,
+  TypeBuilder as __TypeBuilder,
-  t as __t,
+  t as __t,
-  type AlgebraicTypeType as __AlgebraicTypeType,
+  type AlgebraicTypeType as __AlgebraicTypeType,
-  type Infer as __Infer,
+  type Infer as __Infer,
-} from "spacetimedb";
+} from "spacetimedb";
-
+
-export default __t.row({
+export default __t.row({
-  id: __t.u64().primaryKey(),
+  id: __t.u64().primaryKey(),
-  messageId: __t.string().name("message_id"),
+  messageId: __t.string().name("message_id"),
-  userId: __t.string().name("user_id"),
+  userId: __t.string().name("user_id"),
-  userName: __t.string().name("user_name"),
+  userName: __t.string().name("user_name"),
-  reaction: __t.string(),
+  reaction: __t.string(),
-});
+});

web/src/lib/chat/module_bindings/remove_reaction_reducer.ts

@@ -1,18 +1,18 @@
-// THIS FILE IS AUTOMATICALLY GENERATED BY SPACETIMEDB. EDITS TO THIS FILE
+// THIS FILE IS AUTOMATICALLY GENERATED BY SPACETIMEDB. EDITS TO THIS FILE
-// WILL NOT BE SAVED. MODIFY TABLES IN YOUR MODULE SOURCE CODE INSTEAD.
+// WILL NOT BE SAVED. MODIFY TABLES IN YOUR MODULE SOURCE CODE INSTEAD.
-
+
-/* eslint-disable */
+/* eslint-disable */
-/* tslint:disable */
+/* tslint:disable */
-import {
+import {
-  TypeBuilder as __TypeBuilder,
+  TypeBuilder as __TypeBuilder,
-  t as __t,
+  t as __t,
-  type AlgebraicTypeType as __AlgebraicTypeType,
+  type AlgebraicTypeType as __AlgebraicTypeType,
-  type Infer as __Infer,
+  type Infer as __Infer,
-} from "spacetimedb";
+} from "spacetimedb";
-
+
-export default {
+export default {
-  messageId: __t.string(),
+  messageId: __t.string(),
-  workspaceId: __t.string(),
+  workspaceId: __t.string(),
-  userId: __t.string(),
+  userId: __t.string(),
-  reaction: __t.string(),
+  reaction: __t.string(),
-};
+};

web/src/lib/chat/module_bindings/types.ts

@@ -42,3 +42,4 @@ export const SyncOutbox = __t.object("SyncOutbox", {
   synced: __t.bool(),
 });
 export type SyncOutbox = __Infer<typeof SyncOutbox>;
+

web/src/lib/chat/module_bindings/types/reducers.ts

@@ -25,3 +25,4 @@ export type LoadReactionsParams = __Infer<typeof LoadReactionsReducer>;
 export type MarkSyncedParams = __Infer<typeof MarkSyncedReducer>;
 export type RemoveReactionParams = __Infer<typeof RemoveReactionReducer>;
 export type SendMessageParams = __Infer<typeof SendMessageReducer>;
+

web/src/lib/chat/spacetime.svelte.ts

@@ -325,7 +325,7 @@ export function createSpacetimeChat(
 		edit,
 		delete: del,
 		react,
-		refresh: async () => { rebuildMessages(); },
+		refresh: async () => { rebuildMessages(); await enrichFromPg(); },
 		destroy
 	};
 }

web/src/lib/components/ConvertDialog.svelte

@@ -88,7 +88,7 @@
 	<!-- svelte-ignore a11y_click_events_have_key_events -->
 	<!-- svelte-ignore a11y_no_static_element_interactions -->
 	<div class="convert-dialog" onclick={(e) => e.stopPropagation()}>
-		<h3>{type === 'kanban' ? 'Konverter til kanban-kort' : 'Konverter til kalenderhendelse'}</h3>
+		<h3>{type === 'kanban' ? 'Opprett kanban-kort' : 'Opprett kalenderhendelse'}</h3>
 
 		{#if loading}
 			<p class="convert-dialog__loading">Laster...</p>
@@ -147,7 +147,7 @@
 			<button class="convert-dialog__btn convert-dialog__btn--cancel" onclick={onCancel}>Avbryt</button>
 			<button class="convert-dialog__btn convert-dialog__btn--confirm" onclick={handleSubmit}
 				disabled={loading || (type === 'kanban' ? !selectedColumnId : !selectedCalendarId || !startsAt)}
-			>Konverter</button>
+			>Opprett</button>
 		</div>
 	</div>
 </div>

web/src/lib/components/Editor.svelte

@@ -307,27 +307,15 @@
 			<div bind:this={editorEl} class="editor-mount"></div>
 		{/if}
 
-		{#if mode === 'compact'}
+		{#if mode === 'compact' && !expanded}
-			<button type="button" class="expand-btn" onclick={toggleExpand} title={expanded ? 'Minimer' : 'Utvid editor'}>
+			<button type="button" class="expand-btn" onclick={toggleExpand} title="Utvid editor">
-				{#if expanded}
+				<svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
-					<svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+					<polyline points="15 3 21 3 21 9"></polyline>
-						<polyline points="4 14 10 14 10 20"></polyline>
+					<polyline points="9 21 3 21 3 15"></polyline>
-						<polyline points="20 10 14 10 14 4"></polyline>
+					<line x1="21" y1="3" x2="14" y2="10"></line>
-						<line x1="14" y1="10" x2="21" y2="3"></line>
+					<line x1="3" y1="21" x2="10" y2="14"></line>
-						<line x1="3" y1="21" x2="10" y2="14"></line>
+				</svg>
-					</svg>
-				{:else}
-					<svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
-						<polyline points="15 3 21 3 21 9"></polyline>
-						<polyline points="9 21 3 21 3 15"></polyline>
-						<line x1="21" y1="3" x2="14" y2="10"></line>
-						<line x1="3" y1="21" x2="10" y2="14"></line>
-					</svg>
-				{/if}
 			</button>
-		{/if}
-
-		{#if mode === 'compact'}
 			<button
 				type="button"
 				class="send-btn"
@@ -342,6 +330,30 @@
 		{/if}
 	</div>
 
+	{#if mode === 'compact' && expanded}
+		<div class="editor-bottom-bar">
+			<button type="button" class="expand-btn" onclick={toggleExpand} title="Minimer">
+				<svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+					<polyline points="4 14 10 14 10 20"></polyline>
+					<polyline points="20 10 14 10 14 4"></polyline>
+					<line x1="14" y1="10" x2="21" y2="3"></line>
+					<line x1="3" y1="21" x2="10" y2="14"></line>
+				</svg>
+			</button>
+			<button
+				type="button"
+				class="send-btn"
+				onclick={handleSubmit}
+				disabled={!hasContent}
+				aria-label="Send"
+			>
+				<svg width="16" height="16" viewBox="0 0 24 24" fill="currentColor">
+					<path d="M2.01 21L23 12 2.01 3 2 10l15 2-15 2z"/>
+				</svg>
+			</button>
+		</div>
+	{/if}
+
 	<!-- Mention suggestions popup -->
 	{#if suggestions.length > 0 && mentionPopupPos}
 		<div
@@ -616,6 +628,15 @@
 		margin: 0 0.15rem;
 	}
 
+	/* Bottom bar (expanded compact mode) */
+	.editor-bottom-bar {
+		display: flex;
+		justify-content: flex-end;
+		align-items: center;
+		gap: 0.4rem;
+		padding: 0.3rem 0.4rem 0;
+	}
+
 	/* Send button (compact mode) */
 	.send-btn {
 		display: flex;

web/src/lib/components/MessageBox.svelte

@@ -412,8 +412,9 @@
 						{/if}
 					{/if}
 					{#if message.metadata?.ai_action}
-						{@const promptInfo = aiPrompts.find(p => p.action === message.metadata?.ai_action)}
+						<span class="messagebox__ai-badge">
-						<span class="messagebox__ai-badge">{promptInfo?.icon ?? '✨'} {promptInfo?.label ?? message.metadata.ai_action}</span>
+							✨ {message.metadata.ai_label ?? message.metadata.ai_action}{#if message.metadata.ai_model}&ensp;·&ensp;{message.metadata.ai_model}{/if}
+						</span>
 					{/if}
 				</div>
 			{/if}

web/src/lib/server/db.ts

@@ -15,10 +15,10 @@ export interface Workspace {
 	settings: Record<string, unknown>;
 }
 
-/** Hent alle workspaces brukeren er medlem av */
+/** Hent alle workspaces brukeren er medlem av, med rolle */
-export async function getUserWorkspaces(userId: string): Promise<Workspace[]> {
+export async function getUserWorkspaces(userId: string): Promise<(Workspace & { role: string })[]> {
-	return sql<Workspace[]>`
+	return sql<(Workspace & { role: string })[]>`
-		SELECT w.id, w.name, w.slug, w.domain, w.settings
+		SELECT w.id, w.name, w.slug, w.domain, w.settings, wm.role::text AS role
 		FROM workspaces w
 		JOIN workspace_members wm ON wm.workspace_id = w.id
 		WHERE wm.user_id = ${userId}

web/src/lib/types/message.ts

@@ -28,6 +28,8 @@ export interface MessageData {
 		ai_processing?: boolean;
 		ai_processed?: boolean;
 		ai_action?: string;
+		ai_label?: string;
+		ai_model?: string;
 	} | null;
 }
 

web/src/routes/api/admin/ai/models/+server.ts

@@ -1,20 +1,13 @@
 import { json, error } from '@sveltejs/kit';
 import type { RequestHandler } from './$types';
-import { env } from '$env/dynamic/private';
+import { sql } from '$lib/server/db';
-
-interface OpenRouterModel {
-	id: string;
-	name: string;
-	context_length: number;
-	pricing: { prompt: string; completion: string };
-	top_provider?: { max_completion_tokens?: number };
-	architecture?: { modality?: string };
-}
 
 export interface CatalogModel {
 	id: string;
 	name: string;
 	provider: string;
+	litellm_prefix: string;
+	api_key_env: string;
 	context_length: number;
 	prompt_price_per_m: number;
 	completion_price_per_m: number;
@@ -22,46 +15,157 @@ export interface CatalogModel {
 	max_completion: number | null;
 }
 
+interface ProviderFetcher {
+	keyEnv: string;
+	label: string;
+	litellmPrefix: string;
+	fetch: (apiKey: string) => Promise<CatalogModel[]>;
+}
+
 let cache: { models: CatalogModel[]; fetched_at: number } | null = null;
 const CACHE_TTL = 60 * 60 * 1000; // 1 time
 
-function toPerMillion(pricePerToken: string): number {
+function toPerMillion(pricePerToken: string | number): number {
-	const n = parseFloat(pricePerToken);
+	const n = typeof pricePerToken === 'string' ? parseFloat(pricePerToken) : pricePerToken;
 	if (isNaN(n)) return 0;
 	return Math.round(n * 1_000_000 * 100) / 100;
 }
 
-export const GET: RequestHandler = async ({ locals }) => {
+// --- Provider-spesifikke hentere ---
-	if (!locals.workspace || !locals.user) error(401);
-
-	const apiKey = env.OPENROUTER_API_KEY;
-	if (!apiKey) {
-		error(500, 'OPENROUTER_API_KEY er ikke konfigurert');
-	}
-
-	if (cache && Date.now() - cache.fetched_at < CACHE_TTL) {
-		return json(cache.models);
-	}
 
+async function fetchOpenRouter(apiKey: string): Promise<CatalogModel[]> {
 	const res = await fetch('https://openrouter.ai/api/v1/models', {
 		headers: { Authorization: `Bearer ${apiKey}` }
 	});
-
+	if (!res.ok) return [];
-	if (!res.ok) {
-		error(502, `OpenRouter returnerte ${res.status}`);
-	}
-
 	const body = await res.json();
-	const models: CatalogModel[] = (body.data as OpenRouterModel[]).map((m) => ({
+	return (body.data ?? []).map((m: any) => ({
 		id: m.id,
-		name: m.name,
+		name: m.name ?? m.id,
 		provider: m.id.split('/')[0],
-		context_length: m.context_length,
+		litellm_prefix: 'openrouter/',
+		api_key_env: 'OPENROUTER_API_KEY',
+		context_length: m.context_length ?? 0,
 		prompt_price_per_m: toPerMillion(m.pricing?.prompt ?? '0'),
 		completion_price_per_m: toPerMillion(m.pricing?.completion ?? '0'),
 		modality: m.architecture?.modality ?? 'text',
 		max_completion: m.top_provider?.max_completion_tokens ?? null
 	}));
+}
+
+async function fetchXai(_apiKey: string): Promise<CatalogModel[]> {
+	// xAI /v1/models krever betalt konto — hardkod kjente modeller
+	// Kilde: https://docs.x.ai/docs/models
+	const models = [
+		{ id: 'grok-4.20-multi-agent-beta-0309', name: 'Grok 4.20 Multi-Agent (beta)', ctx: 131072 },
+		{ id: 'grok-4.20-beta-0309-reasoning', name: 'Grok 4.20 (reasoning, beta)', ctx: 131072 },
+		{ id: 'grok-4.20-beta-0309-non-reasoning', name: 'Grok 4.20 (beta)', ctx: 131072 },
+		{ id: 'grok-4-0709', name: 'Grok 4', ctx: 131072 },
+		{ id: 'grok-4-fast-reasoning', name: 'Grok 4 Fast (reasoning)', ctx: 131072 },
+		{ id: 'grok-4-fast-non-reasoning', name: 'Grok 4 Fast', ctx: 131072 },
+		{ id: 'grok-4-1-fast-reasoning', name: 'Grok 4.1 Fast (reasoning)', ctx: 131072 },
+		{ id: 'grok-4-1-fast-non-reasoning', name: 'Grok 4.1 Fast', ctx: 131072 },
+		{ id: 'grok-3', name: 'Grok 3', ctx: 131072 },
+		{ id: 'grok-3-mini', name: 'Grok 3 Mini', ctx: 131072 },
+		{ id: 'grok-code-fast-1', name: 'Grok Code Fast', ctx: 131072 },
+	];
+	return models.map(m => ({
+		id: m.id,
+		name: m.name,
+		provider: 'xai',
+		litellm_prefix: 'xai/',
+		api_key_env: 'XAI_API_KEY',
+		context_length: m.ctx,
+		prompt_price_per_m: -1,
+		completion_price_per_m: -1,
+		modality: 'text',
+		max_completion: null
+	}));
+}
+
+async function fetchGemini(apiKey: string): Promise<CatalogModel[]> {
+	const res = await fetch(
+		`https://generativelanguage.googleapis.com/v1beta/models?key=${apiKey}`
+	);
+	if (!res.ok) return [];
+	const body = await res.json();
+	return (body.models ?? [])
+		.filter((m: any) => m.supportedGenerationMethods?.includes('generateContent'))
+		.map((m: any) => {
+			// models/gemini-2.5-flash → gemini-2.5-flash
+			const shortName = (m.name as string).replace('models/', '');
+			return {
+				id: shortName,
+				name: m.displayName ?? shortName,
+				provider: 'google',
+				litellm_prefix: 'gemini/',
+				api_key_env: 'GEMINI_API_KEY',
+				context_length: m.inputTokenLimit ?? 0,
+				prompt_price_per_m: -1,
+				completion_price_per_m: -1,
+				modality: 'text',
+				max_completion: m.outputTokenLimit ?? null
+			};
+		});
+}
+
+async function fetchOpenAI(apiKey: string): Promise<CatalogModel[]> {
+	const res = await fetch('https://api.openai.com/v1/models', {
+		headers: { Authorization: `Bearer ${apiKey}` }
+	});
+	if (!res.ok) return [];
+	const body = await res.json();
+	return (body.data ?? [])
+		.filter((m: any) => m.id.startsWith('gpt-') || m.id.startsWith('o') || m.id.startsWith('chatgpt-'))
+		.map((m: any) => ({
+			id: m.id,
+			name: m.id,
+			provider: 'openai',
+			litellm_prefix: 'openai/',
+			api_key_env: 'OPENAI_API_KEY',
+			context_length: 128000,
+			prompt_price_per_m: 0,
+			completion_price_per_m: 0,
+			modality: 'text',
+			max_completion: null
+		}));
+}
+
+const PROVIDERS: ProviderFetcher[] = [
+	{ keyEnv: 'XAI_API_KEY', label: 'xAI', litellmPrefix: 'xai/', fetch: fetchXai },
+	{ keyEnv: 'GEMINI_API_KEY', label: 'Google', litellmPrefix: 'gemini/', fetch: fetchGemini },
+	{ keyEnv: 'OPENAI_API_KEY', label: 'OpenAI', litellmPrefix: 'openai/', fetch: fetchOpenAI },
+	{ keyEnv: 'OPENROUTER_API_KEY', label: 'OpenRouter', litellmPrefix: 'openrouter/', fetch: fetchOpenRouter }
+];
+
+export const GET: RequestHandler = async ({ locals, url }) => {
+	if (!locals.workspace || !locals.user) error(401);
+
+	const forceRefresh = url.searchParams.get('refresh') === '1';
+
+	if (!forceRefresh && cache && Date.now() - cache.fetched_at < CACHE_TTL) {
+		return json(cache.models);
+	}
+
+	// Hent aktive nøkler med verdier fra DB
+	const keys = await sql`
+		SELECT env_name, key_value FROM ai_api_keys WHERE is_enabled = true AND key_value IS NOT NULL
+	`;
+	const keyMap = new Map(keys.map((k: any) => [k.env_name, k.key_value as string]));
+
+	// Hent fra alle aktive leverandører parallelt
+	const promises = PROVIDERS
+		.filter(p => keyMap.has(p.keyEnv))
+		.map(async (p) => {
+			try {
+				return await p.fetch(keyMap.get(p.keyEnv)!);
+			} catch {
+				return [];
+			}
+		});
+
+	const results = await Promise.all(promises);
+	const models = results.flat();
 
 	cache = { models, fetched_at: Date.now() };
 	return json(models);

web/src/routes/api/admin/ai/prompts/+server.ts

@@ -0,0 +1,16 @@
+import { json, error } from '@sveltejs/kit';
+import type { RequestHandler } from './$types';
+import { sql } from '$lib/server/db';
+
+/** GET — list alle AI-prompts */
+export const GET: RequestHandler = async ({ locals }) => {
+	if (!locals.workspace || !locals.user) error(401);
+
+	const rows = await sql`
+		SELECT action, system_prompt, description, updated_at
+		FROM ai_prompts
+		ORDER BY action
+	`;
+
+	return json(rows);
+};

web/src/routes/api/admin/ai/prompts/[action]/+server.ts

@@ -0,0 +1,44 @@
+import { json, error } from '@sveltejs/kit';
+import type { RequestHandler } from './$types';
+import { sql } from '$lib/server/db';
+
+/** PATCH — oppdater system_prompt, description, label og/eller icon */
+export const PATCH: RequestHandler = async ({ params, request, locals }) => {
+	if (!locals.workspace || !locals.user) error(401);
+
+	const body = await request.json();
+
+	const [row] = await sql`
+		UPDATE ai_prompts SET
+			system_prompt = COALESCE(${body.system_prompt ?? null}, system_prompt),
+			description = COALESCE(${body.description ?? null}, description),
+			label = COALESCE(${body.label ?? null}, label),
+			icon = COALESCE(${body.icon ?? null}, icon),
+			updated_at = now()
+		WHERE action = ${params.action}
+		RETURNING action, system_prompt, description, label, icon, sort_order, updated_at
+	`;
+
+	if (!row) error(404, 'Prompt ikke funnet');
+	return json(row);
+};
+
+/** PUT — opprett eller erstatt prompt for en action */
+export const PUT: RequestHandler = async ({ params, request, locals }) => {
+	if (!locals.workspace || !locals.user) error(401);
+
+	const body = await request.json();
+	if (!body.system_prompt) error(400, 'system_prompt er påkrevd');
+
+	const [row] = await sql`
+		INSERT INTO ai_prompts (action, system_prompt, description)
+		VALUES (${params.action}, ${body.system_prompt}, ${body.description ?? null})
+		ON CONFLICT (action) DO UPDATE SET
+			system_prompt = EXCLUDED.system_prompt,
+			description = COALESCE(EXCLUDED.description, ai_prompts.description),
+			updated_at = now()
+		RETURNING action, system_prompt, description, updated_at
+	`;
+
+	return json(row);
+};

web/src/routes/api/admin/ai/providers/renumber/+server.ts

@@ -0,0 +1,19 @@
+import { json, error } from '@sveltejs/kit';
+import type { RequestHandler } from './$types';
+import { sql } from '$lib/server/db';
+
+/** POST — renummerer prioriteter for en liste providers */
+export const POST: RequestHandler = async ({ request, locals }) => {
+	if (!locals.workspace || !locals.user) error(401);
+
+	const items: { id: string; priority: number }[] = await request.json();
+	if (!Array.isArray(items) || items.length === 0) error(400);
+
+	for (const item of items) {
+		await sql`
+			UPDATE ai_model_providers SET priority = ${item.priority} WHERE id = ${item.id}::uuid
+		`;
+	}
+
+	return json({ ok: true });
+};

web/src/routes/api/channels/[id]/messages/+server.ts

@@ -22,7 +22,7 @@ export const GET: RequestHandler = async ({ params, url, locals }) => {
 	const rows = after
 		? await sql`
 			SELECT m.id, m.channel_id, m.body, m.message_type, m.title,
-				m.pinned, m.visibility, m.created_at, m.updated_at, m.reply_to,
+				m.pinned, m.visibility, m.created_at, m.updated_at, m.reply_to, m.metadata,
 				u.display_name as author_name, u.authentik_id as author_id,
 				(SELECT count(*)::int FROM messages r WHERE r.reply_to = m.id) as reply_count,
 				pm.body as parent_body, pu.display_name as parent_author_name,
@@ -44,7 +44,7 @@ export const GET: RequestHandler = async ({ params, url, locals }) => {
 		`
 		: await sql`
 			SELECT m.id, m.channel_id, m.body, m.message_type, m.title,
-				m.pinned, m.visibility, m.created_at, m.updated_at, m.reply_to,
+				m.pinned, m.visibility, m.created_at, m.updated_at, m.reply_to, m.metadata,
 				u.display_name as author_name, u.authentik_id as author_id,
 				(SELECT count(*)::int FROM messages r WHERE r.reply_to = m.id) as reply_count,
 				pm.body as parent_body, pu.display_name as parent_author_name,
@@ -117,7 +117,8 @@ export const GET: RequestHandler = async ({ params, url, locals }) => {
 			ends_at: m.cal_ends_at ?? null,
 			all_day: m.cal_all_day ?? false,
 			color: m.cal_color ?? null
-		} : null
+		} : null,
+		metadata: m.metadata ?? null
 	}));
 
 	return json(messages);

web/src/routes/api/jobs/[jobId]/+server.ts

@@ -0,0 +1,28 @@
+import { json, error } from '@sveltejs/kit';
+import type { RequestHandler } from './$types';
+import { sql } from '$lib/server/db';
+
+/**
+ * GET /api/jobs/:jobId — Hent status for en jobb.
+ * Workspace-scopet: jobben må tilhøre brukerens workspace.
+ */
+export const GET: RequestHandler = async ({ params, locals }) => {
+	if (!locals.workspace || !locals.user) error(401);
+
+	const [job] = await sql`
+		SELECT id, status, result, error_msg, created_at, completed_at
+		FROM job_queue
+		WHERE id = ${params.jobId}::uuid AND workspace_id = ${locals.workspace.id}
+	`;
+
+	if (!job) error(404, 'Jobb ikke funnet');
+
+	return json({
+		id: job.id,
+		status: job.status,
+		result: job.result,
+		error: job.error_msg,
+		created_at: job.created_at,
+		completed_at: job.completed_at
+	});
+};

web/src/routes/api/messages/[messageId]/reactions/+server.ts

@@ -30,10 +30,15 @@ export const POST: RequestHandler = async ({ params, request, locals }) => {
 	`;
 	if (!msg) error(404, 'Melding ikke funnet');
 
+	// Fjern eventuell eksisterende reaksjon fra denne brukeren (én reaksjon per bruker per melding)
+	await sql`
+		DELETE FROM message_reactions
+		WHERE message_id = ${messageId} AND user_id = ${userId}
+	`;
+
 	await sql`
 		INSERT INTO message_reactions (message_id, user_id, reaction)
 		VALUES (${messageId}, ${userId}, ${reaction})
-		ON CONFLICT DO NOTHING
 	`;
 
 	const reactions = await getReactions(messageId, userId);

web/src/routes/server-admin/ai/+page.server.ts

@@ -22,22 +22,23 @@ export const load: PageServerLoad = async () => {
 	`;
 
 	const prompts = await sql`
-		SELECT action, system_prompt, description, updated_at
+		SELECT action, system_prompt, description, label, icon, sort_order, updated_at
 		FROM ai_prompts
-		ORDER BY action
+		ORDER BY sort_order, action
 	`;
 
 	const usage = await sql`
 		SELECT
 			model_alias,
 			model_actual,
+			action,
 			count(*)::int AS call_count,
 			sum(prompt_tokens)::int AS prompt_tokens,
 			sum(completion_tokens)::int AS completion_tokens,
 			sum(total_tokens)::int AS total_tokens
 		FROM ai_usage_log
 		WHERE created_at > now() - interval '30 days'
-		GROUP BY model_alias, model_actual
+		GROUP BY model_alias, model_actual, action
 		ORDER BY total_tokens DESC
 	`;
 

web/src/routes/server-admin/ai/+page.svelte

@@ -31,12 +31,16 @@
 		action: string;
 		system_prompt: string;
 		description: string | null;
+		label: string | null;
+		icon: string | null;
+		sort_order: number;
 		updated_at: string;
 	}
 
 	interface UsageRow {
 		model_alias: string;
 		model_actual: string | null;
+		action: string | null;
 		call_count: number;
 		prompt_tokens: number;
 		completion_tokens: number;
@@ -47,6 +51,8 @@
 		id: string;
 		name: string;
 		provider: string;
+		litellm_prefix: string;
+		api_key_env: string;
 		context_length: number;
 		prompt_price_per_m: number;
 		completion_price_per_m: number;
@@ -74,6 +80,8 @@
 	let configMsg = $state('');
 	let editingPrompt = $state<string | null>(null);
 	let editPromptText = $state('');
+	let editPromptLabel = $state('');
+	let editPromptIcon = $state('');
 	let expandedAlias = $state<string | null>(null);
 
 	// Alias-redigering
@@ -135,8 +143,9 @@
 		return String(n);
 	}
 
-	function formatPrice(n: number): string {
+	function formatPrice(n: number | null | undefined): string {
-		if (n === 0) return 'Gratis';
+		if (n == null || n < 0) return '\u2014';
+		if (n === 0) return '\u2014';
 		return `$${n.toFixed(2)}`;
 	}
 
@@ -263,6 +272,7 @@
 		}
 	}
 
+	// Grupper etter api_key_env + provider (f.eks. "google via GEMINI_API_KEY" vs "google via OPENROUTER_API_KEY")
 	let groupedByProvider = $derived.by(() => {
 		const search = catalogSearch.toLowerCase();
 		const filtered = search
@@ -270,23 +280,42 @@
 					(m) =>
 						m.name.toLowerCase().includes(search) ||
 						m.id.toLowerCase().includes(search) ||
-						m.provider.toLowerCase().includes(search)
+						m.provider.toLowerCase().includes(search) ||
+						m.api_key_env.toLowerCase().includes(search)
 				)
 			: catalogModels;
 
 		const map = new Map<string, CatalogModel[]>();
 		for (const m of filtered) {
-			const list = map.get(m.provider) ?? [];
+			// Grupper per API-nøkkel, med provider som undergruppe
+			const groupKey = m.api_key_env === 'OPENROUTER_API_KEY'
+				? `${m.provider} (OpenRouter)`
+				: m.provider;
+			const list = map.get(groupKey) ?? [];
 			list.push(m);
-			map.set(m.provider, list);
+			map.set(groupKey, list);
 		}
 
-		// Sorter modeller synkende etter pris innen hver provider
+		// Sorter: pris synkende først (ukjent/-1 sist), deretter navn synkende
 		for (const [, models] of map) {
-			models.sort((a, b) => b.completion_price_per_m - a.completion_price_per_m);
+			models.sort((a, b) => {
+				const aPrice = a.completion_price_per_m;
+				const bPrice = b.completion_price_per_m;
+				const aHasPrice = aPrice > 0;
+				const bHasPrice = bPrice > 0;
+				if (aHasPrice !== bHasPrice) return aHasPrice ? -1 : 1;
+				if (aPrice !== bPrice) return bPrice - aPrice;
+				return b.name.localeCompare(a.name);
+			});
 		}
 
-		return [...map.entries()].sort(([a], [b]) => a.localeCompare(b));
+		// Direkte API-nøkler først, deretter OpenRouter-grupper
+		return [...map.entries()].sort(([a], [b]) => {
+			const aOr = a.includes('OpenRouter');
+			const bOr = b.includes('OpenRouter');
+			if (aOr !== bOr) return aOr ? 1 : -1;
+			return a.localeCompare(b);
+		});
 	});
 
 	let catalogPickerFiltered = $derived.by(() => {
@@ -310,43 +339,11 @@
 		expandedProviders = new Set(expandedProviders);
 	}
 
-	// Mapping fra OpenRouter-provider til LiteLLM direkte-prefiks + nøkkel
+	function litellmModelId(model: CatalogModel): string {
-	const directKeyMap: Record<string, { prefix: string; key: string }> = {
+		return `${model.litellm_prefix}${model.id}`;
-		google:    { prefix: 'gemini/',     key: 'GEMINI_API_KEY' },
-		anthropic: { prefix: 'anthropic/',  key: 'ANTHROPIC_API_KEY' },
-		openai:    { prefix: 'openai/',     key: 'OPENAI_API_KEY' },
-		'x-ai':   { prefix: 'xai/',        key: 'XAI_API_KEY' },
-	};
-
-	function modelForKey(model: CatalogModel, keyEnv: string): string {
-		if (keyEnv === 'OPENROUTER_API_KEY') return `openrouter/${model.id}`;
-		// Direkte: strip provider-prefix fra model.id, legg til LiteLLM-prefiks
-		const mapping = directKeyMap[model.provider];
-		if (mapping) {
-			const modelName = model.id.replace(`${model.provider}/`, '');
-			return `${mapping.prefix}${modelName}`;
-		}
-		return `openrouter/${model.id}`;
 	}
 
-	function availableKeysForModel(model: CatalogModel): ApiKey[] {
+	async function addFromCatalog(model: CatalogModel, aliasId: string) {
-		const keys: ApiKey[] = [];
-		// Direkte nøkkel for denne leverandøren
-		const mapping = directKeyMap[model.provider];
-		if (mapping) {
-			const directKey = apiKeys.find(k => k.name === mapping.key);
-			if (directKey) keys.push(directKey);
-		}
-		// OpenRouter alltid tilgjengelig
-		const orKey = apiKeys.find(k => k.name === 'OPENROUTER_API_KEY');
-		if (orKey) keys.push(orKey);
-		return keys;
-	}
-
-	// Catalog add — steg 1: velg alias, steg 2: velg nøkkel
-	let catalogAddKey = $state('');
-
-	async function addFromCatalog(model: CatalogModel, aliasId: string, keyEnv: string) {
 		errorMsg = '';
 		const maxPri = Math.max(0, ...providersForAlias(aliasId).map((p) => p.priority));
 		try {
@@ -356,8 +353,8 @@
 				body: JSON.stringify({
 					alias_id: aliasId,
 					priority: maxPri + 1,
-					litellm_model: modelForKey(model, keyEnv),
+					litellm_model: litellmModelId(model),
-					api_key_env: keyEnv
+					api_key_env: model.api_key_env
 				})
 			});
 			if (!res.ok) throw new Error('Feil ved opprettelse');
@@ -365,14 +362,14 @@
 			providers = [...providers, row];
 			addingFromCatalog = null;
 			catalogAddAlias = '';
-			catalogAddKey = '';
 		} catch {
 			errorMsg = 'Kunne ikke legge til provider fra katalog';
 		}
 	}
 
 	function selectFromPicker(model: CatalogModel) {
-		newProvider.litellm_model = modelForKey(model, newProvider.api_key_env);
+		newProvider.litellm_model = litellmModelId(model);
+		newProvider.api_key_env = model.api_key_env;
 		showCatalogPicker = false;
 		catalogPickerSearch = '';
 	}
@@ -433,8 +430,12 @@
 
 	function estimateCost(row: UsageRow): number | null {
 		if (!catalogLoaded || !row.model_actual) return null;
-		const model = catalogModels.find((m) => m.id === row.model_actual);
+		// model_actual kan være "xai/grok-..." eller "google/gemma-..." — match mot id eller litellm_prefix+id
-		if (!model) return null;
+		const actual = row.model_actual;
+		const model = catalogModels.find((m) =>
+			m.id === actual || `${m.litellm_prefix}${m.id}` === actual || `${m.provider}/${m.id}` === actual
+		);
+		if (!model || model.prompt_price_per_m < 0 || model.completion_price_per_m < 0) return null;
 		return (
 			(row.prompt_tokens / 1_000_000) * model.prompt_price_per_m +
 			(row.completion_tokens / 1_000_000) * model.completion_price_per_m
@@ -589,11 +590,33 @@
 			const res = await fetch(`/api/admin/ai/providers/${provider.id}`, { method: 'DELETE' });
 			if (!res.ok) throw new Error('Feil');
 			providers = providers.filter((p) => p.id !== provider.id);
+			// Renummerer prioriteter for gjenværende providers under samme alias
+			await renumberPriorities(provider.alias_id);
 		} catch {
 			errorMsg = 'Kunne ikke slette provider';
 		}
 	}
 
+	async function renumberPriorities(aliasId: string) {
+		const ap = providersForAlias(aliasId);
+		let changed = false;
+		for (let i = 0; i < ap.length; i++) {
+			if (ap[i].priority !== i + 1) {
+				ap[i].priority = i + 1;
+				changed = true;
+			}
+		}
+		if (!changed) return;
+		try {
+			await fetch('/api/admin/ai/providers/renumber', {
+				method: 'POST',
+				headers: { 'Content-Type': 'application/json' },
+				body: JSON.stringify(ap.map(p => ({ id: p.id, priority: p.priority })))
+			});
+			providers = [...providers];
+		} catch { /* stille */ }
+	}
+
 	async function addAlias() {
 		errorMsg = '';
 		if (!newAlias.alias) return;
@@ -655,11 +678,15 @@
 	function startEditPrompt(prompt: Prompt) {
 		editingPrompt = prompt.action;
 		editPromptText = prompt.system_prompt;
+		editPromptLabel = prompt.label ?? '';
+		editPromptIcon = prompt.icon ?? '';
 	}
 
 	function cancelEditPrompt() {
 		editingPrompt = null;
 		editPromptText = '';
+		editPromptLabel = '';
+		editPromptIcon = '';
 	}
 
 	async function savePrompt(prompt: Prompt) {
@@ -669,14 +696,22 @@
 			const res = await fetch(`/api/admin/ai/prompts/${prompt.action}`, {
 				method: 'PATCH',
 				headers: { 'Content-Type': 'application/json' },
-				body: JSON.stringify({ system_prompt: editPromptText })
+				body: JSON.stringify({
+					system_prompt: editPromptText,
+					label: editPromptLabel || null,
+					icon: editPromptIcon || null
+				})
 			});
 			if (!res.ok) throw new Error('Feil ved lagring');
 			const updated = await res.json();
 			prompt.system_prompt = updated.system_prompt;
+			prompt.label = updated.label;
+			prompt.icon = updated.icon;
 			prompt.updated_at = updated.updated_at;
 			editingPrompt = null;
 			editPromptText = '';
+			editPromptLabel = '';
+			editPromptIcon = '';
 			markSaved(prompt.action);
 		} catch {
 			errorMsg = 'Kunne ikke lagre prompt';
@@ -816,7 +851,7 @@
 	<!-- Seksjon 1: Modellkatalog -->
 	<section>
 		<div class="catalog-header">
-			<h3>Modellkatalog (OpenRouter)</h3>
+			<h3>Modellkatalog</h3>
 			<div class="catalog-actions">
 				{#if catalogLoaded}
 					<input
@@ -825,6 +860,7 @@
 						placeholder="Søk modeller..."
 						bind:value={catalogSearch}
 					/>
+					<button class="toggle-btn" onclick={() => { catalogLoaded = false; catalogModels = []; }}>Skjul</button>
 				{/if}
 				<button
 					class="toggle-btn"
@@ -867,30 +903,23 @@
 										<span class="cat-col-add">
 											<button
 												class="toggle-btn"
-												onclick={() => { addingFromCatalog = addingFromCatalog === model.id ? null : model.id; catalogAddAlias = ''; catalogAddKey = ''; }}
+												onclick={() => { addingFromCatalog = addingFromCatalog === model.id ? null : model.id; catalogAddAlias = ''; }}
 											>{addingFromCatalog === model.id ? '&#x2717;' : 'Legg til &rarr;'}</button>
 										</span>
 									</div>
 									{#if addingFromCatalog === model.id}
-										{@const modelKeys = availableKeysForModel(model)}
 										<div class="catalog-add-row">
-											<span class="catalog-add-label">Legg til <strong>{model.name}</strong>:</span>
+											<span class="catalog-add-label">Legg til <strong>{model.name}</strong> via {model.api_key_env}:</span>
 											<select bind:value={catalogAddAlias}>
 												<option value="">Velg alias...</option>
 												{#each aliases as a}
 													<option value={a.id}>{a.alias}</option>
 												{/each}
 											</select>
-											<select bind:value={catalogAddKey}>
-												<option value="">Velg nøkkel...</option>
-												{#each modelKeys as k}
-													<option value={k.name}>{k.label}{k.is_enabled ? '' : ' (av)'}</option>
-												{/each}
-											</select>
 											<button
 												class="add-btn"
-												disabled={!catalogAddAlias || !catalogAddKey}
+												disabled={!catalogAddAlias}
-												onclick={() => addFromCatalog(model, catalogAddAlias, catalogAddKey)}
+												onclick={() => addFromCatalog(model, catalogAddAlias)}
 											>Legg til</button>
 										</div>
 									{/if}
@@ -1139,6 +1168,7 @@
 		<div class="table-list">
 			<div class="table-row table-row--header prompt-row">
 				<span class="col-action">Action</span>
+				<span class="col-label">Visningsnavn</span>
 				<span class="col-desc">Beskrivelse</span>
 				<span class="col-chars">Tegn</span>
 				<span class="col-updated">Oppdatert</span>
@@ -1148,6 +1178,7 @@
 			{#each prompts as prompt (prompt.action)}
 				<div class="table-row prompt-row">
 					<span class="col-action">{prompt.action}</span>
+					<span class="col-label">{prompt.icon ?? ''} {prompt.label ?? '\u2014'}</span>
 					<span class="col-desc">{prompt.description ?? '\u2014'}</span>
 					<span class="col-chars">{prompt.system_prompt.length}</span>
 					<span class="col-updated">{new Date(prompt.updated_at).toLocaleDateString('nb-NO')}</span>
@@ -1164,6 +1195,16 @@
 
 				{#if editingPrompt === prompt.action}
 					<div class="prompt-editor">
+						<div class="prompt-editor-meta">
+							<label>
+								<span class="prompt-meta-label">Ikon</span>
+								<input type="text" bind:value={editPromptIcon} placeholder="🧹" class="prompt-meta-input prompt-meta-input--icon" />
+							</label>
+							<label>
+								<span class="prompt-meta-label">Visningsnavn</span>
+								<input type="text" bind:value={editPromptLabel} placeholder="Vask tekst" class="prompt-meta-input" />
+							</label>
+						</div>
 						<textarea
 							bind:value={editPromptText}
 							rows="12"
@@ -1196,8 +1237,9 @@
 				<div class="table-row table-row--header usage-row">
 					<span>Alias</span>
 					<span>Modell</span>
+					<span>Prompt</span>
 					<span class="col-num">Kall</span>
-					<span class="col-num">Prompt</span>
+					<span class="col-num">Prompt-tok.</span>
 					<span class="col-num">Kompl.</span>
 					<span class="col-num">Totalt</span>
 					<span class="col-num">Est. $</span>
@@ -1208,6 +1250,7 @@
 					<div class="table-row usage-row">
 						<span class="col-alias">{row.model_alias}</span>
 						<span class="col-model-actual">{row.model_actual ?? '\u2014'}</span>
+						<span class="col-action-usage">{row.action ?? '\u2014'}</span>
 						<span class="col-num">{row.call_count}</span>
 						<span class="col-num">{row.prompt_tokens.toLocaleString('nb-NO')}</span>
 						<span class="col-num">{row.completion_tokens.toLocaleString('nb-NO')}</span>
@@ -2040,7 +2083,15 @@
 
 	/* Usage/tokenforbruk */
 	.usage-row {
-		grid-template-columns: 1fr 2fr 70px 80px 80px 80px 70px;
+		grid-template-columns: 1fr 2fr 1fr 60px 80px 80px 80px 60px;
+	}
+
+	.col-action-usage {
+		font-size: 0.75rem;
+		color: #8b92a5;
+		overflow: hidden;
+		text-overflow: ellipsis;
+		white-space: nowrap;
 	}
 
 	.col-model-actual {
@@ -2060,7 +2111,11 @@
 
 	/* Prompt-seksjon */
 	.prompt-row {
-		grid-template-columns: 1.5fr 2.5fr 60px 90px 70px;
+		grid-template-columns: 1.2fr 1.2fr 2fr 60px 80px 70px;
+	}
+
+	.col-label {
+		font-size: 0.85rem;
 	}
 
 	.col-action {
@@ -2089,6 +2144,43 @@
 		border-bottom: 1px solid #2d3148;
 	}
 
+	.prompt-editor-meta {
+		display: flex;
+		gap: 0.75rem;
+		margin-bottom: 0.5rem;
+	}
+
+	.prompt-editor-meta label {
+		display: flex;
+		align-items: center;
+		gap: 0.35rem;
+	}
+
+	.prompt-meta-label {
+		font-size: 0.75rem;
+		color: #8b92a5;
+	}
+
+	.prompt-meta-input {
+		background: #161822;
+		border: 1px solid #2d3148;
+		border-radius: 4px;
+		color: #e1e4e8;
+		padding: 0.25rem 0.4rem;
+		font-size: 0.8rem;
+		width: 160px;
+	}
+
+	.prompt-meta-input--icon {
+		width: 50px;
+		text-align: center;
+	}
+
+	.prompt-meta-input:focus {
+		outline: none;
+		border-color: #3b82f6;
+	}
+
 	.prompt-editor textarea {
 		width: 100%;
 		background: #161822;

worker/src/handlers/ai_text_process.rs

@@ -163,28 +163,44 @@ impl JobHandler for AiTextProcessHandler {
         .await
         .context("Feil ved lagring av revisjon")?;
 
-        // 4. Bygg system-prompt basert på action
+        // 4. Hent prompt-label fra DB (for metadata-stempel i chat)
+        let prompt_label: Option<String> = sqlx::query_scalar(
+            "SELECT label FROM ai_prompts WHERE action = $1"
+        )
+        .bind(action)
+        .fetch_optional(pool)
+        .await
+        .ok()
+        .flatten();
+
+        // 5. Bygg system-prompt basert på action
         let system_prompt = match prompt_override {
             Some(custom) => custom.to_string(),
             None => get_system_prompt_from_db(pool, action).await,
         };
 
-        // 5. Send til AI Gateway
+        // 6. Send til AI Gateway
         let ai_resp = self
             .call_ai_gateway(&system_prompt, &plain_text, &model)
             .await
             .context("AI Gateway-kall feilet")?;
 
-        // 6. Oppdater SpacetimeDB — dette er primær-kanalen til frontend.
+        // 7. Beregn faktisk modellnavn
-        //    sync.rs synker body-endringen til PG automatisk.
+        // LiteLLM returnerer alias-navnet i model-feltet — bruk expected_model fra DB
-        self.update_spacetimedb(&message_id, workspace_id, &ai_resp.content)
+        let actual_model = match &ai_resp.model_actual {
-            .await
+            Some(m) if m != &model => Some(m.clone()),  // Gateway returnerte faktisk modellnavn
-            .context("Kunne ikke oppdatere SpacetimeDB med AI-resultat")?;
+            _ => expected_model,                         // Bruk oppslaget fra providers-tabellen
+        };
+        // Strip openrouter/-prefiks for lesbarhet
+        let actual_model_clean = actual_model.map(|m| m.replace("openrouter/", "").replace("gemini/", "google/"));
 
-        // 7. Skriv PG-only data (metadata, revisjon er allerede lagret, tokenforbruk)
+        // 8. Skriv PG metadata FØR SpacetimeDB-oppdatering
+        //    (frontend henter metadata fra PG når SpacetimeDB-update trigger onUpdate)
         let metadata = json!({
             "ai_processed": true,
-            "ai_action": action
+            "ai_action": action,
+            "ai_label": prompt_label.as_deref().unwrap_or(action),
+            "ai_model": actual_model_clean.as_deref().unwrap_or(&model)
         });
 
         sqlx::query(
@@ -194,31 +210,30 @@ impl JobHandler for AiTextProcessHandler {
             WHERE id = $2
             "#,
         )
-        .bind(metadata)
+        .bind(&metadata)
         .bind(message_id)
         .execute(pool)
         .await
         .context("Feil ved oppdatering av metadata")?;
 
-        // 8. Logg tokenforbruk til ai_usage_log
+        // 9. Oppdater SpacetimeDB — dette er primær-kanalen til frontend.
-        // LiteLLM returnerer alias-navnet i model-feltet — bruk expected_model fra DB
+        //    Gjøres ETTER PG-metadata slik at enrichMessageFromPg finner fersk data.
-        let actual_model = match &ai_resp.model_actual {
+        self.update_spacetimedb(&message_id, workspace_id, &ai_resp.content)
-            Some(m) if m != &model => Some(m.clone()),  // Gateway returnerte faktisk modellnavn
+            .await
-            _ => expected_model,                         // Bruk oppslaget fra providers-tabellen
+            .context("Kunne ikke oppdatere SpacetimeDB med AI-resultat")?;
-        };
-        // Strip openrouter/-prefiks for lesbarhet
-        let actual_model_clean = actual_model.map(|m| m.replace("openrouter/", "").replace("gemini/", "google/"));
 
+        // 10. Logg tokenforbruk til ai_usage_log
         sqlx::query(
             r#"
-            INSERT INTO ai_usage_log (workspace_id, job_id, job_type, model_alias, model_actual, prompt_tokens, completion_tokens, total_tokens)
+            INSERT INTO ai_usage_log (workspace_id, job_id, job_type, model_alias, model_actual, action, prompt_tokens, completion_tokens, total_tokens)
-            VALUES ($1, $2, 'ai_text_process', $3, $4, $5, $6, $7)
+            VALUES ($1, $2, 'ai_text_process', $3, $4, $5, $6, $7, $8)
             "#,
         )
         .bind(workspace_id)
         .bind(job_id)
         .bind(&model)
         .bind(&actual_model_clean)
+        .bind(action)
         .bind(ai_resp.prompt_tokens)
         .bind(ai_resp.completion_tokens)
         .bind(ai_resp.total_tokens)