Skip to content

LLM Wiki is a cross-platform desktop application that turns your documents into an organized, interlinked knowledge base — automatically. Instead of traditional RAG (retrieve-and-answer from scratch every time), the LLM incrementally builds and maintains a persistent wiki from your sources。

TypeScript

Harden Azure OpenAI provider support

nash_sufe605c2931eb
diff --git a/src/components/settings/llm-presets.ts b/src/components/settings/llm-presets.ts
index 58e06d11..72fe4898 100644
--- a/src/components/settings/llm-presets.ts
+++ b/src/components/settings/llm-presets.ts
@@ -1,3 +1,5 @@
+import type { AzureModelFamily } from "@/stores/wiki-store"
+
/**
* Curated LLM provider presets.
*
@@ -42,6 +44,8 @@ export interface LlmPreset {
defaultModel?: string
/** Azure OpenAI api-version query parameter. Azure deployments vary by resource. */
azureApiVersion?: string
+ /** Azure deployment names are arbitrary, so users can declare GPT-5/o-series behavior explicitly. */
+ azureModelFamily?: AzureModelFamily
/**
* Curated list of model ids the UI shows as clickable chips above the
* Model input. The user can still type a custom value — the input stays
diff --git a/src/components/settings/preset-resolver.test.ts b/src/components/settings/preset-resolver.test.ts
index f8140150..c911aa09 100644
--- a/src/components/settings/preset-resolver.test.ts
+++ b/src/components/settings/preset-resolver.test.ts
@@ -50,4 +50,24 @@ describe("resolveConfig", () => {
expect(resolved.reasoning).toEqual({ mode: "off" })
})
+
+ it("carries Azure API version and model family overrides", () => {
+ const preset: LlmPreset = {
+ id: "azure",
+ label: "Azure OpenAI",
+ provider: "azure",
+ baseUrl: "https://resource.openai.azure.com",
+ defaultModel: "wiki-main",
+ azureApiVersion: "2024-10-21",
+ }
+
+ const resolved = resolveConfig(
+ preset,
+ { azureApiVersion: "2025-01-01-preview", azureModelFamily: "gpt5" },
+ fallbackConfig(),
+ )
+
+ expect(resolved.azureApiVersion).toBe("2025-01-01-preview")
+ expect(resolved.azureModelFamily).toBe("gpt5")
+ })
})
diff --git a/src/components/settings/preset-resolver.ts b/src/components/settings/preset-resolver.ts
index d1dbf1f8..dc26ba82 100644
--- a/src/components/settings/preset-resolver.ts
+++ b/src/components/settings/preset-resolver.ts
@@ -53,6 +53,7 @@ export function resolveConfig(
ollamaUrl: fallback.ollamaUrl,
customEndpoint: ov.baseUrl ?? preset.baseUrl ?? "",
azureApiVersion: ov.azureApiVersion ?? preset.azureApiVersion ?? AZURE_OPENAI_API_VERSION,
+ azureModelFamily: ov.azureModelFamily ?? preset.azureModelFamily ?? "auto",
maxContextSize,
reasoning,
}
diff --git a/src/components/settings/sections/llm-provider-section.tsx b/src/components/settings/sections/llm-provider-section.tsx
index 44544a63..bd8c86a8 100644
--- a/src/components/settings/sections/llm-provider-section.tsx
+++ b/src/components/settings/sections/llm-provider-section.tsx
@@ -125,10 +125,11 @@ function PresetRow({
const apiMode = ov.apiMode ?? preset.apiMode ?? "chat_completions"
const baseUrl = ov.baseUrl ?? preset.baseUrl ?? ""
const azureApiVersion = ov.azureApiVersion ?? preset.azureApiVersion ?? AZURE_OPENAI_API_VERSION
+ const azureModelFamily = ov.azureModelFamily ?? preset.azureModelFamily ?? "auto"
const context = ov.maxContextSize ?? preset.suggestedContextSize ?? 131072
const reasoning = ov.reasoning ?? { mode: "auto" as const }
const [testState, setTestState] = useState<ProviderTestState>({ kind: "idle" })
- const hasConfig = !!apiKey || !!ov.baseUrl || !!ov.model || !!ov.azureApiVersion
+ const hasConfig = !!apiKey || !!ov.baseUrl || !!ov.model || !!ov.azureApiVersion || !!ov.azureModelFamily
// Local CLI providers authenticate via their own existing login state
// (inherited by the spawned subprocess), so no API key field is shown.
// Ollama ditto for its local-only model.
@@ -139,7 +140,7 @@ function PresetRow({
const resolvedConfig = useMemo(
() => resolveConfig(preset, ov, useWikiStore.getState().llmConfig),
- [apiKey, apiMode, azureApiVersion, baseUrl, context, model, preset, reasoning, ov],
+ [apiKey, apiMode, azureApiVersion, azureModelFamily, baseUrl, context, model, preset, reasoning, ov],
)
async function runProviderTest(kind: "connection" | "function") {
@@ -277,16 +278,32 @@ function PresetRow({
)}
{preset.provider === "azure" && (
- <div className="space-y-2">
- <Label>Azure API version</Label>
- <Input
- value={azureApiVersion}
- onChange={(e) => onChange({ azureApiVersion: e.target.value })}
- placeholder="2024-10-21"
- />
- <p className="text-xs text-muted-foreground">
- Sent as the Azure OpenAI <code className="font-mono">api-version</code> query parameter.
- </p>
+ <div className="grid gap-4 md:grid-cols-2">
+ <div className="space-y-2">
+ <Label>{t("settings.sections.llm.azureApiVersion")}</Label>
+ <Input
+ value={azureApiVersion}
+ onChange={(e) => onChange({ azureApiVersion: e.target.value })}
+ placeholder="2024-10-21"
+ />
+ <p className="text-xs text-muted-foreground">
+ {t("settings.sections.llm.azureApiVersionHint")}
+ </p>
+ </div>
+ <div className="space-y-2">
+ <Label>{t("settings.sections.llm.azureModelFamily")}</Label>
+ <select
+ className="w-full rounded-md border bg-background px-3 py-2 text-sm focus:outline-none focus:ring-2 focus:ring-ring"
+ value={azureModelFamily}
+ onChange={(e) => onChange({ azureModelFamily: e.target.value as typeof azureModelFamily })}
+ >
+ <option value="auto">{t("settings.sections.llm.azureModelFamilyAuto")}</option>
+ <option value="gpt5">{t("settings.sections.llm.azureModelFamilyGpt5")}</option>
+ </select>
+ <p className="text-xs text-muted-foreground">
+ {t("settings.sections.llm.azureModelFamilyHint")}
+ </p>
+ </div>
</div>
)}
diff --git a/src/components/settings/sections/multimodal-section.tsx b/src/components/settings/sections/multimodal-section.tsx
index 271384e2..308f2d4f 100644
--- a/src/components/settings/sections/multimodal-section.tsx
+++ b/src/components/settings/sections/multimodal-section.tsx
@@ -206,19 +206,32 @@ export function MultimodalSection({ draft, setDraft }: Props) {
)}
{draft.multimodalProvider === "azure" && (
- <div className="space-y-2">
- <Label>{t("settings.sections.multimodal.azureApiVersion", "Azure API version")}</Label>
- <Input
- value={draft.multimodalAzureApiVersion}
- onChange={(e) => setDraft("multimodalAzureApiVersion", e.target.value)}
- placeholder="2024-10-21"
- />
- <p className="text-xs text-muted-foreground">
- {t(
- "settings.sections.multimodal.azureApiVersionHint",
- "Sent as the Azure OpenAI api-version query parameter.",
- )}
- </p>
+ <div className="grid gap-4 md:grid-cols-2">
+ <div className="space-y-2">
+ <Label>{t("settings.sections.multimodal.azureApiVersion")}</Label>
+ <Input
+ value={draft.multimodalAzureApiVersion}
+ onChange={(e) => setDraft("multimodalAzureApiVersion", e.target.value)}
+ placeholder="2024-10-21"
+ />
+ <p className="text-xs text-muted-foreground">
+ {t("settings.sections.multimodal.azureApiVersionHint")}
+ </p>
+ </div>
+ <div className="space-y-2">
+ <Label>{t("settings.sections.multimodal.azureModelFamily")}</Label>
+ <select
+ className="w-full rounded-md border bg-background px-3 py-2 text-sm focus:outline-none focus:ring-2 focus:ring-ring"
+ value={draft.multimodalAzureModelFamily}
+ onChange={(e) => setDraft("multimodalAzureModelFamily", e.target.value as typeof draft.multimodalAzureModelFamily)}
+ >
+ <option value="auto">{t("settings.sections.multimodal.azureModelFamilyAuto")}</option>
+ <option value="gpt5">{t("settings.sections.multimodal.azureModelFamilyGpt5")}</option>
+ </select>
+ <p className="text-xs text-muted-foreground">
+ {t("settings.sections.multimodal.azureModelFamilyHint")}
+ </p>
+ </div>
</div>
)}
diff --git a/src/components/settings/settings-types.ts b/src/components/settings/settings-types.ts
index 5478d959..01397218 100644
--- a/src/components/settings/settings-types.ts
+++ b/src/components/settings/settings-types.ts
@@ -1,5 +1,5 @@
import type { CustomApiMode } from "./llm-presets"
-import type { ReasoningConfig, SourceWatchConfig } from "@/stores/wiki-store"
+import type { AzureModelFamily, ReasoningConfig, SourceWatchConfig } from "@/stores/wiki-store"
/**
* Shape of the draft state each section reads from and writes into.
@@ -15,6 +15,7 @@ export interface SettingsDraft {
ollamaUrl: string
customEndpoint: string
azureApiVersion: string
+ azureModelFamily: AzureModelFamily
maxContextSize: number
apiMode: CustomApiMode | undefined
reasoning: ReasoningConfig | undefined
@@ -40,6 +41,7 @@ export interface SettingsDraft {
multimodalOllamaUrl: string
multimodalCustomEndpoint: string
multimodalAzureApiVersion: string
+ multimodalAzureModelFamily: AzureModelFamily
multimodalApiMode: CustomApiMode | undefined
multimodalConcurrency: number
diff --git a/src/components/settings/settings-view.tsx b/src/components/settings/settings-view.tsx
index 4f7c104a..5a8f13c0 100644
--- a/src/components/settings/settings-view.tsx
+++ b/src/components/settings/settings-view.tsx
@@ -109,6 +109,7 @@ function initialDraft(
ollamaUrl: llm.ollamaUrl,
customEndpoint: llm.customEndpoint,
azureApiVersion: llm.azureApiVersion ?? "2024-10-21",
+ azureModelFamily: llm.azureModelFamily ?? "auto",
maxContextSize: llm.maxContextSize ?? 204800,
apiMode: llm.apiMode,
reasoning: llm.reasoning,
@@ -127,6 +128,7 @@ function initialDraft(
multimodalOllamaUrl: multimodal.ollamaUrl,
multimodalCustomEndpoint: multimodal.customEndpoint,
multimodalAzureApiVersion: multimodal.azureApiVersion ?? "2024-10-21",
+ multimodalAzureModelFamily: multimodal.azureModelFamily ?? "auto",
multimodalApiMode: multimodal.apiMode,
multimodalConcurrency: multimodal.concurrency,
outputLanguage,
@@ -272,6 +274,7 @@ export function SettingsView() {
ollamaUrl: draft.ollamaUrl,
customEndpoint: draft.customEndpoint,
azureApiVersion: draft.provider === "azure" ? draft.azureApiVersion.trim() : undefined,
+ azureModelFamily: draft.provider === "azure" ? draft.azureModelFamily : undefined,
maxContextSize: draft.maxContextSize,
apiMode: draft.provider === "custom" ? draft.apiMode : undefined,
reasoning: draft.reasoning,
@@ -294,6 +297,7 @@ export function SettingsView() {
ollamaUrl: draft.multimodalOllamaUrl,
customEndpoint: draft.multimodalCustomEndpoint,
azureApiVersion: draft.multimodalProvider === "azure" ? draft.multimodalAzureApiVersion.trim() : undefined,
+ azureModelFamily: draft.multimodalProvider === "azure" ? draft.multimodalAzureModelFamily : undefined,
apiMode: draft.multimodalProvider === "custom" ? draft.multimodalApiMode : undefined,
// Clamp at save time so a hand-edited persisted store with a
// ridiculous concurrency value (e.g. someone setting 1000 in
diff --git a/src/i18n/en.json b/src/i18n/en.json
index 8ea36a08..a61ca6e5 100644
--- a/src/i18n/en.json
+++ b/src/i18n/en.json
@@ -223,6 +223,13 @@
"testFunction": "Test function",
"testingConnection": "Testing connection…",
"testingFunction": "Testing function…",
+ "deploymentName": "Deployment name",
+ "azureApiVersion": "Azure API version",
+ "azureApiVersionHint": "Sent as the Azure OpenAI api-version query parameter.",
+ "azureModelFamily": "Azure model family",
+ "azureModelFamilyAuto": "Auto-detect from deployment name",
+ "azureModelFamilyGpt5": "GPT-5 / o-series strict parameters",
+ "azureModelFamilyHint": "Azure deployment names are arbitrary. Choose GPT-5 / o-series if this deployment rejects max_tokens or non-default temperature.",
"reasoning": {
"title": "Reasoning / thinking",
"auto": "Auto",
@@ -284,6 +291,15 @@
"ollamaUrl": "Ollama URL",
"customEndpoint": "Endpoint URL",
"customEndpointHint": "OpenAI-compatible /v1 base. LM Studio, llama.cpp server, vLLM, LocalAI all work.",
+ "azureEndpoint": "Azure endpoint",
+ "azureEndpointHint": "Use your Azure OpenAI resource endpoint. The model field is the deployment name.",
+ "azureApiVersion": "Azure API version",
+ "azureApiVersionHint": "Sent as the Azure OpenAI api-version query parameter.",
+ "azureModelFamily": "Azure model family",
+ "azureModelFamilyAuto": "Auto-detect from deployment name",
+ "azureModelFamilyGpt5": "GPT-5 / o-series strict parameters",
+ "azureModelFamilyHint": "Choose GPT-5 / o-series if this deployment rejects max_tokens or non-default temperature.",
+ "azureDeployment": "Deployment name",
"apiKey": "API key",
"apiKeyPlaceholder": "Leave blank for local / no-auth endpoints",
"model": "Model",
diff --git a/src/i18n/zh.json b/src/i18n/zh.json
index b78cadcf..1e4d09e1 100644
--- a/src/i18n/zh.json
+++ b/src/i18n/zh.json
@@ -223,6 +223,13 @@
"testFunction": "测试功能",
"testingConnection": "正在测试连接…",
"testingFunction": "正在测试功能…",
+ "deploymentName": "部署名称",
+ "azureApiVersion": "Azure API 版本",
+ "azureApiVersionHint": "会作为 Azure OpenAI 的 api-version 查询参数发送。",
+ "azureModelFamily": "Azure 模型族",
+ "azureModelFamilyAuto": "根据部署名称自动判断",
+ "azureModelFamilyGpt5": "GPT-5 / o-series 严格参数",
+ "azureModelFamilyHint": "Azure 的 deployment 名称可以任意命名。如果该部署拒绝 max_tokens 或非默认 temperature,请选择 GPT-5 / o-series。",
"reasoning": {
"title": "Reasoning / thinking",
"auto": "自动",
@@ -284,6 +291,15 @@
"ollamaUrl": "Ollama URL",
"customEndpoint": "端点 URL",
"customEndpointHint": "OpenAI 兼容的 /v1 根地址。LM Studio、llama.cpp server、vLLM、LocalAI 都可以。",
+ "azureEndpoint": "Azure 端点",
+ "azureEndpointHint": "填写 Azure OpenAI resource endpoint。模型字段填写 deployment name。",
+ "azureApiVersion": "Azure API 版本",
+ "azureApiVersionHint": "会作为 Azure OpenAI 的 api-version 查询参数发送。",
+ "azureModelFamily": "Azure 模型族",
+ "azureModelFamilyAuto": "根据部署名称自动判断",
+ "azureModelFamilyGpt5": "GPT-5 / o-series 严格参数",
+ "azureModelFamilyHint": "如果该部署拒绝 max_tokens 或非默认 temperature,请选择 GPT-5 / o-series。",
+ "azureDeployment": "部署名称",
"apiKey": "API Key",
"apiKeyPlaceholder": "本地 / 无鉴权端点留空",
"model": "模型",
diff --git a/src/lib/azure-openai.test.ts b/src/lib/azure-openai.test.ts
index 0d470ab7..144db4e6 100644
--- a/src/lib/azure-openai.test.ts
+++ b/src/lib/azure-openai.test.ts
@@ -1,5 +1,5 @@
import { describe, it, expect } from "vitest"
-import { buildAzureOpenAiUrl } from "./azure-openai"
+import { buildAzureOpenAiUrl, isAzureOpenAiEndpoint } from "./azure-openai"
describe("buildAzureOpenAiUrl", () => {
it("builds deployment chat URL with api-version", () => {
@@ -25,4 +25,11 @@ describe("buildAzureOpenAiUrl", () => {
"https://my-resource.openai.azure.com/openai/deployments/my-gpt5/chat/completions?api-version=2024-10-21",
)
})
+
+ it("detects only real Azure OpenAI hostnames", () => {
+ expect(isAzureOpenAiEndpoint("https://my-resource.openai.azure.com")).toBe(true)
+ expect(isAzureOpenAiEndpoint("my-resource.openai.azure.com/openai/deployments/wiki")).toBe(true)
+ expect(isAzureOpenAiEndpoint("https://my-resource.openai.azure.com.evil.com")).toBe(false)
+ expect(isAzureOpenAiEndpoint("https://example.com/path/openai.azure.com")).toBe(false)
+ })
})
diff --git a/src/lib/azure-openai.ts b/src/lib/azure-openai.ts
index 881b0ffd..983d2647 100644
--- a/src/lib/azure-openai.ts
+++ b/src/lib/azure-openai.ts
@@ -1,7 +1,14 @@
export const AZURE_OPENAI_API_VERSION = "2024-10-21"
export function isAzureOpenAiEndpoint(endpoint: string): boolean {
- return /\.openai\.azure\.com/i.test(endpoint)
+ const trimmed = endpoint.trim()
+ if (!trimmed) return false
+ try {
+ const url = new URL(/^https?:\/\//i.test(trimmed) ? trimmed : `https://${trimmed}`)
+ return url.hostname.toLowerCase().endsWith(".openai.azure.com")
+ } catch {
+ return /(^|\/\/)[^/?#]+\.openai\.azure\.com(?::\d+)?(?:[/?#]|$)/i.test(trimmed)
+ }
}
export interface AzureParsedEndpoint {
diff --git a/src/lib/ingest.ts b/src/lib/ingest.ts
index 47a6c815..a64616a6 100644
--- a/src/lib/ingest.ts
+++ b/src/lib/ingest.ts
@@ -42,6 +42,8 @@ function resolveCaptionConfig(
model: mm.model,
ollamaUrl: mm.ollamaUrl,
customEndpoint: mm.customEndpoint,
+ azureApiVersion: mm.azureApiVersion,
+ azureModelFamily: mm.azureModelFamily,
apiMode: mm.apiMode,
// The caption helper hits `streamChat` directly, which doesn't
// care about `maxContextSize` (that field is for the analysis
diff --git a/src/lib/llm-providers.test.ts b/src/lib/llm-providers.test.ts
index de7375c6..9b25a73c 100644
--- a/src/lib/llm-providers.test.ts
+++ b/src/lib/llm-providers.test.ts
@@ -193,6 +193,40 @@ describe("Azure OpenAI provider", () => {
expect(body.temperature).toBeUndefined()
})
+
+ it("uses explicit Azure model family when deployment name does not reveal GPT-5", () => {
+ const cfg = getProviderConfig(mkConfig({
+ provider: "azure",
+ model: "production-chat-deployment",
+ customEndpoint: "https://example-resource.openai.azure.com",
+ azureModelFamily: "gpt5",
+ }))
+ const body = cfg.buildBody(
+ [{ role: "user", content: "hi" }],
+ { max_tokens: 1024, temperature: 0.1 },
+ ) as Record<string, unknown>
+
+ expect(body.max_tokens).toBeUndefined()
+ expect(body.max_completion_tokens).toBe(1024)
+ expect(body.temperature).toBeUndefined()
+ })
+
+ it("applies explicit Azure model family to custom Azure endpoints", () => {
+ const cfg = getProviderConfig(mkConfig({
+ provider: "custom",
+ model: "wiki-main",
+ customEndpoint: "https://example-resource.openai.azure.com",
+ azureModelFamily: "gpt5",
+ }))
+ const body = cfg.buildBody(
+ [{ role: "user", content: "hi" }],
+ { max_tokens: 512, temperature: 0.1 },
+ ) as Record<string, unknown>
+
+ expect(body.model).toBeUndefined()
+ expect(body.max_completion_tokens).toBe(512)
+ expect(body.temperature).toBeUndefined()
+ })
})
describe("Ollama / custom (chat_completions) — vision content", () => {
diff --git a/src/lib/llm-providers.ts b/src/lib/llm-providers.ts
index 1ea9455a..1478bf18 100644
--- a/src/lib/llm-providers.ts
+++ b/src/lib/llm-providers.ts
@@ -251,6 +251,11 @@ function isKimiEndpoint(config: LlmConfig): boolean {
}
function isOpenAiStrictCompletionModel(config: LlmConfig): boolean {
+ if ((config.provider === "azure" || (config.provider === "custom" && isAzureOpenAiEndpoint(config.customEndpoint)))
+ && config.azureModelFamily === "gpt5") {
+ return true
+ }
+
const model = config.model.trim().toLowerCase()
const strictModel =
/^gpt-5(?:[.\-_]|$)/.test(model) || /^o\d+(?:[.\-_]|$)/.test(model)
diff --git a/src/stores/wiki-store.ts b/src/stores/wiki-store.ts
index d6775147..4303a18c 100644
--- a/src/stores/wiki-store.ts
+++ b/src/stores/wiki-store.ts
@@ -9,6 +9,7 @@ import { DEFAULT_SOURCE_WATCH_CONFIG } from "@/lib/source-watch-config"
* `chat_completions` for backward compatibility with pre-0.3.7 configs.
*/
export type CustomApiMode = "chat_completions" | "anthropic_messages"
+export type AzureModelFamily = "auto" | "gpt5"
export type ReasoningMode = "auto" | "off" | "low" | "medium" | "high" | "max" | "custom"
export interface ReasoningConfig {
@@ -23,6 +24,7 @@ interface LlmConfig {
ollamaUrl: string
customEndpoint: string
azureApiVersion?: string
+ azureModelFamily?: AzureModelFamily
maxContextSize: number // max context window in characters
apiMode?: CustomApiMode
reasoning?: ReasoningConfig
@@ -186,6 +188,7 @@ interface MultimodalConfig {
ollamaUrl: string
customEndpoint: string
azureApiVersion?: string
+ azureModelFamily?: AzureModelFamily
apiMode?: CustomApiMode
/** Max parallel caption requests during ingest. >=1. */
concurrency: number
@@ -231,6 +234,7 @@ export interface ProviderOverride {
model?: string
baseUrl?: string // customEndpoint for custom presets, ollamaUrl for ollama
azureApiVersion?: string
+ azureModelFamily?: AzureModelFamily
apiMode?: CustomApiMode
maxContextSize?: number
reasoning?: ReasoningConfig