feat(ai): UI-configurable response timeout, local streaming, dedicated OCR provider

Add three per-user AI settings, all configured in the app (no new env vars),
behind one migration (0198):

- Response timeout (seconds). A per-user upstream timeout threaded onto the
  provider call. A local/self-hosted backend can take more than the legacy
  60 s default to load a model on the first request, which timed the Coach
  out before the first token. Defaults to a generous ~180 s when unset;
  validated to 10–600 s.

- True token streaming for the local OpenAI-compatible client. The client now
  POSTs with stream:true and parses the SSE delta chunks, emitting tokens as
  they arrive, with a per-idle-gap timeout instead of a whole-call ceiling and
  a graceful fallback to the buffered path when a server cannot stream. The
  Coach route runs the provider call inside the response stream behind a
  keepalive heartbeat so a reverse proxy no longer idle-drops the connection
  during model load or generation. Grounding, refusal, and sentinel-stripping
  still run on the complete reply before any token reaches the client.

- Dedicated document-scan (Lab-OCR) provider. Off by default — the OCR path
  uses the main provider chain. When enabled, lab-report processing resolves a
  separate provider/model/key, with a hint when the chosen model is not
  vision-capable. The new key column is AES-256-GCM at rest and registered in
  the encrypted-columns registry and the rotation script.

i18n: EN + DE translations for the new Settings → AI strings (other locales
carry interim copies to keep locale parity green).
This commit is contained in:
Marc-André Bombeck committed 2026-06-27 15:21:40 +02:00
1 parent 3b74b108a1
commit b91649d6de
30 files changed
+2281 -569

No files matched your search

@@ -31,9 +31,11 @@ import { queryKeys } from "@/lib/query-keys";
import { AdminOpenAIProviderForm } from "./admin-openai-provider-form";
import { AnthropicProviderForm } from "./anthropic-provider-form";
import { CodexProviderForm } from "./codex-provider-form";
import { DocumentScanCard } from "./document-scan-card";
import { FallbackChainCard } from "./fallback-chain-card";
import { LocalProviderForm } from "./local-provider-form";
import { OpenAIProviderForm } from "./openai-provider-form";
import { ResponseTimeoutCard } from "./response-timeout-card";
import { RuntimeActionsRow } from "./runtime-actions-row";
import {
PROVIDER_TYPES,
@@ -171,6 +173,13 @@ export function AiInsightsCard({
onSelect={pickProvider}
/>
{/* v1.22 (#89) — per-user response timeout (mainly for slow
local/self-hosted backends). */}
<ResponseTimeoutCard userProvider={userProvider} />
{/* v1.22 (#90) — dedicated document-scan (Lab-OCR) provider. */}
<DocumentScanCard userProvider={userProvider} />
<RuntimeActionsRow
provider={selectedProvider}
userProvider={userProvider}
@@ -0,0 +1,212 @@
"use client";
/* ────────────────────────────────────────────────────────────────
* v1.22 (#90) — Document scanning (Lab reports).
*
* Opt-in dedicated provider for the Lab-OCR ingestion path. OFF by default:
* scanning uses the main provider chain. When enabled, document/lab-report
* processing routes to THIS provider/model/key instead — so a user can keep a
* cheap text Coach on one provider and send expensive vision OCR to a
* vision-capable one. Surfaces a hint when the chosen model can't read images.
* ──────────────────────────────────────────────────────────────── */
import { useState } from "react";
import { useMutation, useQueryClient } from "@tanstack/react-query";
import { Loader2, Save } from "lucide-react";
import { Button } from "@/components/ui/button";
import { Input } from "@/components/ui/input";
import { Label } from "@/components/ui/label";
import { NativeSelect } from "@/components/ui/native-select";
import { PasswordInput } from "@/components/ui/password-input";
import { Switch } from "@/components/ui/switch";
import { apiPatch } from "@/lib/api/api-fetch";
import {
supportsVisionForConfig,
type VisionProviderType,
} from "@/lib/ai/vision-capability";
import { useTranslations } from "@/lib/i18n/context";
import { queryKeys } from "@/lib/query-keys";
import { OCR_PROVIDER_TYPES, type UserAIProvider } from "./shared";
/** Map the OCR provider enum to the vision-capability tag. */
function toVisionType(provider: string): VisionProviderType {
switch (provider) {
case "ANTHROPIC":
return "anthropic";
case "OPENAI":
return "openai";
case "LOCAL":
return "local";
default:
return "none";
}
}
export function DocumentScanCard({
userProvider,
}: {
userProvider: UserAIProvider | null | undefined;
}) {
const { t } = useTranslations();
const queryClient = useQueryClient();
const [enabled, setEnabled] = useState(false);
const [provider, setProvider] = useState<string>("OPENAI");
const [model, setModel] = useState("");
const [baseUrl, setBaseUrl] = useState("");
const [apiKey, setApiKey] = useState("");
const [msg, setMsg] = useState<string | null>(null);
const [ok, setOk] = useState(false);
const seededKey =
userProvider != null
? `${userProvider.ocrEnabled}|${userProvider.ocrProvider ?? ""}|${userProvider.ocrModel ?? ""}|${userProvider.ocrBaseUrl ?? ""}`
: null;
const [previousSeed, setPreviousSeed] = useState<string | null>(null);
if (seededKey != null && seededKey !== previousSeed) {
setPreviousSeed(seededKey);
setEnabled(userProvider?.ocrEnabled ?? false);
setProvider(userProvider?.ocrProvider ?? "OPENAI");
setModel(userProvider?.ocrModel ?? "");
setBaseUrl(userProvider?.ocrBaseUrl ?? "");
}
// Vision-capability hint: warn when a model is chosen that can't read images
// (local is trust-by-default; the warning surfaces only for cloud providers).
const visionOk =
!enabled ||
model.trim() === "" ||
supportsVisionForConfig(toVisionType(provider), model.trim());
const saveMutation = useMutation({
mutationFn: async () => {
const body: Record<string, unknown> = {
ocrEnabled: enabled,
ocrProvider: provider,
ocrModel: model.trim() || null,
ocrBaseUrl: provider === "LOCAL" ? baseUrl.trim() || null : null,
};
if (apiKey.trim()) body.ocrKey = apiKey.trim();
await apiPatch("/api/user/ai-provider", body);
},
onSuccess: () => {
setOk(true);
setMsg(t("settings.ai.saved"));
setApiKey("");
queryClient.invalidateQueries({ queryKey: queryKeys.userAiProvider() });
},
onError: (e) => {
setOk(false);
setMsg(e instanceof Error ? e.message : t("settings.ai.errorGeneric"));
},
});
return (
<div className="bg-muted/50 space-y-4 rounded-lg p-4">
<div className="flex items-start justify-between gap-3">
<div>
<p className="text-sm font-medium">{t("settings.ai.ocrHeading")}</p>
<p className="text-muted-foreground text-xs">
{t("settings.ai.ocrBody")}
</p>
</div>
<Switch
checked={enabled}
onCheckedChange={setEnabled}
aria-label={t("settings.ai.ocrEnableLabel")}
data-testid="ai-ocr-enable"
/>
</div>
{enabled && (
<div className="space-y-4">
<div>
<Label htmlFor="ai-ocr-provider">
{t("settings.ai.ocrProviderLabel")}
</Label>
<NativeSelect
id="ai-ocr-provider"
value={provider}
onChange={(e) => setProvider(e.target.value)}
className="mt-1 sm:max-w-md"
>
{OCR_PROVIDER_TYPES.map((p) => (
<option key={p} value={p}>
{t(`settings.ai.ocrProviderOption.${p}` as const)}
</option>
))}
</NativeSelect>
</div>
<div>
<Label htmlFor="ai-ocr-model">{t("settings.ai.modelLabel")}</Label>
<Input
id="ai-ocr-model"
value={model}
onChange={(e) => setModel(e.target.value)}
placeholder={t("settings.ai.ocrModelPlaceholder")}
className="mt-1"
/>
{!visionOk && (
<p className="text-warning mt-1 text-xs">
{t("settings.ai.ocrVisionHint")}
</p>
)}
</div>
{provider === "LOCAL" && (
<div>
<Label htmlFor="ai-ocr-base-url">
{t("settings.ai.baseUrlLabel")}
</Label>
<Input
id="ai-ocr-base-url"
value={baseUrl}
onChange={(e) => setBaseUrl(e.target.value)}
placeholder="http://localhost:11434/v1"
className="mt-1"
/>
</div>
)}
<div>
<Label htmlFor="ai-ocr-key">{t("settings.ai.ocrKeyLabel")}</Label>
<PasswordInput
id="ai-ocr-key"
value={apiKey}
onChange={(e) => setApiKey(e.target.value)}
placeholder={
userProvider?.hasOcrKey ? t("settings.ai.savedShort") : ""
}
className="mt-1"
/>
</div>
</div>
)}
<div>
<Button
size="sm"
className="min-h-11 sm:min-h-9"
onClick={() => saveMutation.mutate()}
disabled={saveMutation.isPending}
>
{saveMutation.isPending ? (
<Loader2 className="h-4 w-4 animate-spin motion-reduce:animate-none" />
) : (
<Save className="h-4 w-4" />
)}
{t("settings.ai.saveCta")}
</Button>
</div>
{msg && (
<p className={`text-xs ${ok ? "text-success" : "text-destructive"}`}>
{msg}
</p>
)}
</div>
);
}
@@ -0,0 +1,126 @@
"use client";
/* ────────────────────────────────────────────────────────────────
* v1.22 (#89) — Response timeout (seconds).
*
* A per-user upstream timeout for AI generation, threaded onto the provider
* call (`CompletionParams.timeoutMs`). Surfaced here mainly for local /
* self-hosted backends: an MLX/exo server can take >60 s on the first request
* while it loads the model, which the legacy 60 s default timed out before the
* first token landed. Empty = the generous built-in default (~180 s).
* ──────────────────────────────────────────────────────────────── */
import { useState } from "react";
import { useMutation, useQueryClient } from "@tanstack/react-query";
import { Loader2, Save } from "lucide-react";
import { Button } from "@/components/ui/button";
import { Input } from "@/components/ui/input";
import { Label } from "@/components/ui/label";
import { apiPatch } from "@/lib/api/api-fetch";
import { useTranslations } from "@/lib/i18n/context";
import { queryKeys } from "@/lib/query-keys";
import type { UserAIProvider } from "./shared";
export function ResponseTimeoutCard({
userProvider,
}: {
userProvider: UserAIProvider | null | undefined;
}) {
const { t } = useTranslations();
const queryClient = useQueryClient();
const [value, setValue] = useState("");
const [msg, setMsg] = useState<string | null>(null);
const [ok, setOk] = useState(false);
// Seed from the persisted value once it arrives (seed-on-data pattern).
const seededKey =
userProvider != null
? `${userProvider.responseTimeoutSeconds ?? ""}`
: null;
const [previousSeed, setPreviousSeed] = useState<string | null>(null);
if (seededKey != null && seededKey !== previousSeed) {
setPreviousSeed(seededKey);
setValue(
userProvider?.responseTimeoutSeconds != null
? String(userProvider.responseTimeoutSeconds)
: "",
);
}
const saveMutation = useMutation({
mutationFn: async () => {
const trimmed = value.trim();
const responseTimeoutSeconds = trimmed === "" ? null : Number(trimmed);
if (
responseTimeoutSeconds !== null &&
(!Number.isInteger(responseTimeoutSeconds) ||
responseTimeoutSeconds < 10 ||
responseTimeoutSeconds > 600)
) {
throw new Error(t("settings.ai.responseTimeoutInvalid"));
}
await apiPatch("/api/user/ai-provider", { responseTimeoutSeconds });
},
onSuccess: () => {
setOk(true);
setMsg(t("settings.ai.saved"));
queryClient.invalidateQueries({ queryKey: queryKeys.userAiProvider() });
},
onError: (e) => {
setOk(false);
setMsg(e instanceof Error ? e.message : t("settings.ai.errorGeneric"));
},
});
return (
<div className="bg-muted/50 space-y-3 rounded-lg p-4">
<div>
<p className="text-sm font-medium">
{t("settings.ai.responseTimeoutHeading")}
</p>
<p className="text-muted-foreground text-xs">
{t("settings.ai.responseTimeoutBody")}
</p>
</div>
<div>
<Label htmlFor="ai-response-timeout">
{t("settings.ai.responseTimeoutLabel")}
</Label>
<Input
id="ai-response-timeout"
type="number"
inputMode="numeric"
min={10}
max={600}
value={value}
onChange={(e) => setValue(e.target.value)}
placeholder={t("settings.ai.responseTimeoutPlaceholder")}
className="mt-1 sm:max-w-xs"
/>
</div>
<div>
<Button
size="sm"
className="min-h-11 sm:min-h-9"
onClick={() => saveMutation.mutate()}
disabled={saveMutation.isPending}
>
{saveMutation.isPending ? (
<Loader2 className="h-4 w-4 animate-spin motion-reduce:animate-none" />
) : (
<Save className="h-4 w-4" />
)}
{t("settings.ai.saveCta")}
</Button>
</div>
{msg && (
<p className={`text-xs ${ok ? "text-success" : "text-destructive"}`}>
{msg}
</p>
)}
</div>
);
}
+13
View File
@@ -27,8 +27,21 @@ export interface UserAIProvider {
hasLocalKey: boolean;
hasOpenaiKey: boolean;
openaiKeyPreview: string | null;
// v1.22 (#89) — per-user response timeout, in seconds (null = default).
responseTimeoutSeconds: number | null;
// v1.22 (#90) — dedicated document-scan (Lab-OCR) provider config.
ocrEnabled: boolean;
ocrProvider: string | null;
ocrModel: string | null;
ocrBaseUrl: string | null;
hasOcrKey: boolean;
ocrKeyPreview: string | null;
}
/** v1.22 (#90) — providers the dedicated document-scan path accepts. */
export const OCR_PROVIDER_TYPES = ["OPENAI", "ANTHROPIC", "LOCAL"] as const;
export type OcrProviderType = (typeof OCR_PROVIDER_TYPES)[number];
/**
* v1.4.16 phase B2 — provider tags exposed to the UI. Mirrors
* `PROVIDER_CHAIN_TYPES` server-side; kept as a const-array here to