feat(provider): add LiteLLM proxy provider

README.md

@@ -415,6 +415,10 @@ Common commands include `/start`, `/help`, `/pair`, `/new`, `/sessions`, `/use`,
       <img src="./src/renderer/src/assets/llm-icons/requesty.png" width="50" height="50" alt="Requesty Icon"><br/>
       <a href="https://www.requesty.ai/">Requesty</a>
     </td>
+    <td>
+      <img src="./src/renderer/src/assets/llm-icons/litellm.svg" width="50" height="50" alt="LiteLLM Icon"><br/>
+      <a href="https://www.litellm.ai/">LiteLLM</a>
+    </td>
     <td>
       <img src="./src/renderer/src/assets/llm-icons/opper.svg" width="50" height="50" alt="Opper Icon"><br/>
       <a href="https://opper.ai/">Opper</a>

src/main/provider/defaults.ts

@@ -580,6 +580,20 @@ const PROVIDER_DEFAULTS: LLM_PROVIDER_BASE[] = [
       models: 'https://www.requesty.ai/models'
     }
   },
+  {
+    id: 'litellm',
+    name: 'LiteLLM',
+    apiType: 'openai-completions',
+    apiKey: '',
+    baseUrl: 'http://localhost:4000/v1',
+    enable: false,
+    websites: {
+      official: 'https://www.litellm.ai/',
+      apiKey: 'https://docs.litellm.ai/docs/proxy/virtual_keys',
+      docs: 'https://docs.litellm.ai/docs/simple_proxy',
+      models: 'https://docs.litellm.ai/docs/providers'
+    }
+  },
   {
     id: 'opper',
     name: 'Opper',

src/main/provider/providerImportService.ts

@@ -125,6 +125,7 @@ const PROVIDER_ID_ALIASES: Record<string, string> = {
   ppinfra: 'ppio',
   openrouter: 'openrouter',
   requesty: 'requesty',
+  litellm: 'litellm',
   demonroute: 'demonroute',
   orcarouter: 'orcarouter',
   synthorai: 'synthorai',

src/main/provider/providerRegistry.ts

@@ -24,6 +24,7 @@ export type AiSdkModelSourceStrategy =
   | 'new-api'
   | 'openrouter'
   | 'requesty'
+  | 'litellm'
   | 'demonroute'
   | 'ppio'
   | 'groq'
@@ -34,6 +35,7 @@ export type AiSdkModelSourceStrategy =
 
 export type AiSdkKeyStatusStrategy =
   | 'none'
+  | 'litellm'
   | 'openrouter'
   | 'deepseek'
   | 'ppio'
@@ -594,6 +596,15 @@ const PROVIDER_ID_REGISTRY = new Map<string, AiSdkProviderDefinition>([
       credentialStrategy: 'api-key'
     })
   ],
+  [
+    'litellm',
+    // Self-hosted proxy: the key is optional (master or virtual key only when auth is on).
+    createDefinition({
+      ...OPENAI_BASE,
+      modelSource: 'litellm',
+      keyStatusStrategy: 'litellm'
+    })
+  ],
   [
     'demonroute',
     createDefinition({

src/main/provider/providers/aiSdkProvider.ts

@@ -187,6 +187,24 @@ function toModelRecordArray(payload: unknown): Array<Record<string, unknown>> {
   return []
 }
 
+// LiteLLM `model_info.mode` -> the DeepChat model type whose route the proxy also serves.
+// `responses` aliases are served on chat completions too.
+const LITELLM_MODE_TYPES: Record<string, ModelType | undefined> = {
+  chat: ModelType.Chat,
+  responses: ModelType.Chat,
+  embedding: ModelType.Embedding,
+  image_generation: ModelType.ImageGeneration,
+  video_generation: ModelType.VideoGeneration,
+  audio_speech: ModelType.TTS
+}
+
+// LiteLLM `litellm_provider` -> model group, e.g. `vertex_ai-language-models` -> `vertex_ai` and
+// `text-completion-codestral` -> `codestral`.
+function toLiteLLMProviderGroup(value: unknown): string {
+  if (typeof value !== 'string' || !value) return 'default'
+  return value.replace(/^text-completion-/i, '').split('-')[0] || 'default'
+}
+
 function toPositiveFiniteNumber(value: unknown): number | undefined {
   return typeof value === 'number' && Number.isFinite(value) && value > 0 ? value : undefined
 }
@@ -1741,6 +1759,8 @@ export class AiSdkProvider extends BaseLLMProvider {
         return this.fetchOpenAiDerivedModels(strategy)
       case 'requesty':
         return this.fetchRequestyModels()
+      case 'litellm':
+        return this.fetchLiteLLMModels()
       case 'demonroute':
         return this.fetchDemonRouteModels()
       case 'bedrock':
@@ -1978,6 +1998,91 @@ export class AiSdkProvider extends BaseLLMProvider {
     return models
   }
 
+  /** LiteLLM management routes (`/model/info`, `/key/info`) live at the proxy root, not `/v1`. */
+  private getLiteLLMProxyRoot(): string {
+    return (this.provider.baseUrl || '').replace(/\/+$/, '').replace(/\/v1$/i, '')
+  }
+
+  /**
+   * LiteLLM model names are aliases chosen by whoever runs the proxy, so the shared model catalog
+   * cannot describe them. The proxy's `/model/info` carries the routed model's mode, limits and
+   * capabilities per alias; `/models` (ids only) is the fallback when the proxy returns no model
+   * records or the key cannot read them (401, 403, 404). Other errors are surfaced.
+   */
+  private async fetchLiteLLMModels(): Promise<MODEL_META[]> {
+    const timeout = this.getModelFetchTimeout()
+    const infoUrl = `${this.getLiteLLMProxyRoot()}/model/info`
+    let records: Array<Record<string, unknown>>
+    try {
+      records = toModelRecordArray(
+        await this.requestProviderJson<unknown>(infoUrl, { method: 'GET' }, timeout)
+      )
+    } catch (error) {
+      if (!(error instanceof ProviderHttpError) || ![401, 403, 404].includes(error.status ?? 0)) {
+        throw error
+      }
+      console.warn('LiteLLM model info unavailable, falling back to /models:', error.status)
+      records = []
+    }
+
+    // Single-model CLI proxies return a data object with a wildcard name; /models has callable IDs.
+    if (records.length === 0) {
+      const listed = await this.fetchDefaultOpenAIModels({ timeout })
+      return listed.filter((model) => !model.id.includes('*'))
+    }
+
+    const models = new Map<string, MODEL_META>()
+    for (const record of records) {
+      const modelId = typeof record.model_name === 'string' ? record.model_name : ''
+      const info =
+        record.model_info && typeof record.model_info === 'object'
+          ? (record.model_info as Record<string, unknown>)
+          : {}
+      // Modes without a DeepChat route (rerank, audio_transcription, moderation, ...) are skipped,
+      // and so are wildcard routes (`*`, `openai/*`), which are patterns rather than callable names.
+      const type = LITELLM_MODE_TYPES[typeof info.mode === 'string' ? info.mode : 'chat']
+      if (!modelId || modelId.includes('*') || !type) {
+        continue
+      }
+      const contextLength = toPositiveFiniteNumber(info.max_input_tokens)
+      const maxTokens = toPositiveFiniteNumber(info.max_output_tokens)
+      const existing = models.get(modelId)
+      if (existing) {
+        // Deployments of one alias are load balanced, so a request can land on any of them: keep
+        // the smallest limits and only the capabilities every deployment reports.
+        if (contextLength !== undefined) {
+          existing.contextLength = Math.min(existing.contextLength ?? contextLength, contextLength)
+        }
+        if (maxTokens !== undefined) {
+          existing.maxTokens = Math.min(existing.maxTokens ?? maxTokens, maxTokens)
+        }
+        existing.vision = existing.vision === true && info.supports_vision === true
+        existing.functionCall =
+          existing.functionCall === true && info.supports_function_calling === true
+        existing.reasoning = existing.reasoning === true && info.supports_reasoning === true
+        continue
+      }
+      models.set(modelId, {
+        id: modelId,
+        name: modelId,
+        group: toLiteLLMProviderGroup(info.litellm_provider),
+        providerId: this.provider.id,
+        isCustom: false,
+        type,
+        ...(contextLength !== undefined ? { contextLength } : {}),
+        ...(maxTokens !== undefined ? { maxTokens } : {}),
+        ...(typeof info.supports_vision === 'boolean' ? { vision: info.supports_vision } : {}),
+        ...(typeof info.supports_function_calling === 'boolean'
+          ? { functionCall: info.supports_function_calling }
+          : {}),
+        ...(typeof info.supports_reasoning === 'boolean'
+          ? { reasoning: info.supports_reasoning }
+          : {})
+      })
+    }
+    return [...models.values()]
+  }
+
   /**
    * DemonRoute exposes a standard OpenAI-compatible model catalog at `{baseUrl}/models`.
    * The catalog request is authenticated and doubles as the key check.
@@ -2250,6 +2355,43 @@ export class AiSdkProvider extends BaseLLMProvider {
         }
         return keyStatus
       }
+      case 'litellm': {
+        // Spend and budget exist for virtual keys only; a keyless proxy or the master key has none.
+        if (!this.provider.apiKey) return null
+        const { signal, dispose } = this.createModelRequestSignal({
+          timeout: this.getModelFetchTimeout()
+        })
+        try {
+          const response = await this.fetchProvider(`${this.getLiteLLMProxyRoot()}/key/info`, {
+            method: 'GET',
+            headers: { Authorization: `Bearer ${this.provider.apiKey}` },
+            signal
+          })
+          if (response.status === 404) return null
+          if (!response.ok) {
+            const errorText = await response.text()
+            throw new Error(
+              `LiteLLM key check failed: ${response.status} ${response.statusText} - ${errorText}`
+            )
+          }
+          const payload = (await response.json()) as {
+            info?: { spend?: number | null; max_budget?: number | null }
+          }
+          // Per-request spend is often below a cent; keep enough digits to show it.
+          const usd = (value: number) => `$${value.toFixed(Math.abs(value) < 0.01 ? 6 : 2)}`
+          const spend = Number(payload.info?.spend ?? 0)
+          const keyStatus: KeyStatus = { usage: usd(spend) }
+          const budget = payload.info?.max_budget
+          if (typeof budget === 'number') {
+            const remaining = budget - spend
+            keyStatus.limit_remaining = usd(remaining)
+            keyStatus.remainNum = remaining
+          }
+          return keyStatus
+        } finally {
+          dispose()
+        }
+      }
       case 'deepseek': {
         const response = await this.fetchProvider('https://api.deepseek.com/user/balance', {
           method: 'GET',

src/renderer/settings/components/ProviderApiConfig.vue

@@ -487,7 +487,7 @@ watch(
     keyStatusError.value = ''
     if (
       !storedApiKey ||
-      !['ppio', 'openrouter', 'deepseek', '302ai', 'cherryin'].includes(providerId)
+      !['ppio', 'openrouter', 'deepseek', '302ai', 'cherryin', 'litellm'].includes(providerId)
     )
       return
     try {

src/renderer/src/components/icons/modelIconRegistry.ts

@@ -50,6 +50,7 @@ import kimiColorIcon from '@/assets/llm-icons/kimi-color.svg?url'
 import moonshotColorIcon from '@/assets/llm-icons/moonshot.svg?url'
 import openrouterColorIcon from '@/assets/llm-icons/openrouter.svg?url'
 import requestyColorIcon from '@/assets/llm-icons/requesty.png?url'
+import litellmColorIcon from '@/assets/llm-icons/litellm.svg?url'
 import opperColorIcon from '@/assets/llm-icons/opper.svg?url'
 import demonrouteColorIcon from '@/assets/llm-icons/demonroute.svg?url'
 import orcarouterColorIcon from '@/assets/llm-icons/orcarouter.svg?url'
@@ -189,6 +190,7 @@ export const modelIcons = {
   moonshot: moonshotColorIcon,
   openrouter: openrouterColorIcon,
   requesty: requestyColorIcon,
+  litellm: litellmColorIcon,
   opper: opperColorIcon,
   demonroute: demonrouteColorIcon,
   orcarouter: orcarouterColorIcon,