Feature: Add local OpenAI-compatible model discovery to /model (#201)

* Add local OpenAI-compatible model discovery to /model * Guard local OpenAI model discovery from Codex routing * Preserve remote OpenAI Codex alias behavior
2026-04-05 15:46:06 -07:00
parent 60d3d8961a
commit c534aa5771
15 changed files with 539 additions and 39 deletions
--- a/src/utils/config.ts
+++ b/src/utils/config.ts
@@ -576,6 +576,7 @@ export type GlobalConfig = {

  // Additional model options for the model picker (fetched during bootstrap).
  additionalModelOptionsCache?: ModelOption[]
+  additionalModelOptionsCacheScope?: string

  // Additional model options discovered from OpenAI-compatible endpoints.
  openaiAdditionalModelOptionsCache?: ModelOption[]
--- a/src/utils/model/modelOptions.ts
+++ b/src/utils/model/modelOptions.ts
@@ -1,5 +1,6 @@
 // biome-ignore-all assist/source/organizeImports: internal-only import markers must not be reordered
 import { getInitialMainLoopModel } from '../../bootstrap/state.js'
+import { getAdditionalModelOptionsCacheScope } from '../../services/api/providerConfig.js'
 import {
  isClaudeAISubscriber,
  isMaxSubscriber,
@@ -44,6 +45,25 @@ export type ModelOption = {
  descriptionForModel?: string
 }

+function getScopedAdditionalModelOptions(): ModelOption[] {
+  const config = getGlobalConfig()
+  const activeScope = getAdditionalModelOptionsCacheScope()
+
+  if (!activeScope) {
+    return []
+  }
+
+  if (config.additionalModelOptionsCacheScope !== undefined) {
+    return config.additionalModelOptionsCacheScope === activeScope
+      ? (config.additionalModelOptionsCache ?? [])
+      : []
+  }
+
+  return activeScope === 'firstParty'
+    ? (config.additionalModelOptionsCache ?? [])
+    : []
+}
+
 export function getDefaultOptionForUser(fastMode = false): ModelOption {
  if (process.env.USER_TYPE === 'ant') {
    const currentModel = renderDefaultModelSetting(
@@ -408,6 +428,16 @@ function getModelOptionsBase(fastMode = false): ModelOption[] {
    return standardOptions
  }

+  if (getAdditionalModelOptionsCacheScope()?.startsWith('openai:')) {
+    const activeOpenAIOptions = getActiveOpenAIModelOptionsCache()
+    return [
+      getDefaultOptionForUser(fastMode),
+      ...(activeOpenAIOptions.length > 0
+        ? activeOpenAIOptions
+        : getScopedAdditionalModelOptions()),
+    ]
+  }
+
  // PAYG 1P API: Default (Sonnet) + Sonnet 1M + Opus 4.6 + Opus 1M + Haiku
  if (getAPIProvider() === 'firstParty') {
    const payg1POptions = [getDefaultOptionForUser(fastMode)]
@@ -566,13 +596,8 @@ export function getModelOptions(fastMode = false): ModelOption[] {
    })
  }

-  const additionalOptions =
-    getAPIProvider() === 'openai'
-      ? getActiveOpenAIModelOptionsCache()
-      : getGlobalConfig().additionalModelOptionsCache ?? []
-
-  // Append additional model options fetched during bootstrap/endpoints.
-  for (const opt of additionalOptions) {
+  // Append additional model options fetched during bootstrap
+  for (const opt of getScopedAdditionalModelOptions()) {
    if (!options.some(existing => existing.value === opt.value)) {
      options.push(opt)
    }
--- a/src/utils/model/providers.test.ts
+++ b/src/utils/model/providers.test.ts
@@ -7,6 +7,9 @@ const originalEnv = {
  CLAUDE_CODE_USE_BEDROCK: process.env.CLAUDE_CODE_USE_BEDROCK,
  CLAUDE_CODE_USE_VERTEX: process.env.CLAUDE_CODE_USE_VERTEX,
  CLAUDE_CODE_USE_FOUNDRY: process.env.CLAUDE_CODE_USE_FOUNDRY,
+  OPENAI_BASE_URL: process.env.OPENAI_BASE_URL,
+  OPENAI_API_BASE: process.env.OPENAI_API_BASE,
+  OPENAI_MODEL: process.env.OPENAI_MODEL,
 }

 afterEach(() => {
@@ -16,6 +19,9 @@ afterEach(() => {
  process.env.CLAUDE_CODE_USE_BEDROCK = originalEnv.CLAUDE_CODE_USE_BEDROCK
  process.env.CLAUDE_CODE_USE_VERTEX = originalEnv.CLAUDE_CODE_USE_VERTEX
  process.env.CLAUDE_CODE_USE_FOUNDRY = originalEnv.CLAUDE_CODE_USE_FOUNDRY
+  process.env.OPENAI_BASE_URL = originalEnv.OPENAI_BASE_URL
+  process.env.OPENAI_API_BASE = originalEnv.OPENAI_API_BASE
+  process.env.OPENAI_MODEL = originalEnv.OPENAI_MODEL
 })

 async function importFreshProvidersModule() {
@@ -29,6 +35,9 @@ function clearProviderEnv(): void {
  delete process.env.CLAUDE_CODE_USE_BEDROCK
  delete process.env.CLAUDE_CODE_USE_VERTEX
  delete process.env.CLAUDE_CODE_USE_FOUNDRY
+  delete process.env.OPENAI_BASE_URL
+  delete process.env.OPENAI_API_BASE
+  delete process.env.OPENAI_MODEL
 }

 test('first-party provider keeps Anthropic account setup flow enabled', () => {
@@ -69,3 +78,29 @@ test('GEMINI takes precedence over GitHub when both are set', async () => {

  expect(getAPIProvider()).toBe('gemini')
 })
+
+test('explicit local openai-compatible base URLs stay on the openai provider', () => {
+  clearProviderEnv()
+  process.env.CLAUDE_CODE_USE_OPENAI = '1'
+  process.env.OPENAI_BASE_URL = 'http://127.0.0.1:8080/v1'
+  process.env.OPENAI_MODEL = 'gpt-5.4'
+
+  expect(getAPIProvider()).toBe('openai')
+})
+
+test('codex aliases still resolve to the codex provider without a non-codex base URL', () => {
+  clearProviderEnv()
+  process.env.CLAUDE_CODE_USE_OPENAI = '1'
+  process.env.OPENAI_MODEL = 'codexplan'
+
+  expect(getAPIProvider()).toBe('codex')
+})
+
+test('official OpenAI base URLs now keep provider detection on openai for aliases', () => {
+  clearProviderEnv()
+  process.env.CLAUDE_CODE_USE_OPENAI = '1'
+  process.env.OPENAI_BASE_URL = 'https://api.openai.com/v1'
+  process.env.OPENAI_MODEL = 'gpt-5.4'
+
+  expect(getAPIProvider()).toBe('openai')
+})
--- a/src/utils/model/providers.ts
+++ b/src/utils/model/providers.ts
@@ -1,5 +1,5 @@
 import type { AnalyticsMetadata_I_VERIFIED_THIS_IS_NOT_CODE_OR_FILEPATHS } from '../../services/analytics/index.js'
-import { isCodexAlias } from '../../services/api/providerConfig.js'
+import { shouldUseCodexTransport } from '../../services/api/providerConfig.js'
 import { isEnvTruthy } from '../envUtils.js'

 export type APIProvider =
@@ -34,11 +34,10 @@ export function usesAnthropicAccountFlow(): boolean {
  return getAPIProvider() === 'firstParty'
 }
 function isCodexModel(): boolean {
-  const model = (process.env.OPENAI_MODEL || '').trim()
-  if (!model) return false
-  // Delegate to the canonical alias table in providerConfig to keep
-  // the two Codex detection systems (provider type + transport) in sync.
-  return isCodexAlias(model)
+  return shouldUseCodexTransport(
+    process.env.OPENAI_MODEL || '',
+    process.env.OPENAI_BASE_URL ?? process.env.OPENAI_API_BASE,
+  )
 }

 export function getAPIProviderForStatsig(): AnalyticsMetadata_I_VERIFIED_THIS_IS_NOT_CODE_OR_FILEPATHS {
--- a/src/utils/providerDiscovery.test.ts
+++ b/src/utils/providerDiscovery.test.ts
@@ -0,0 +1,78 @@
+import { afterEach, expect, mock, test } from 'bun:test'
+
+import {
+  getLocalOpenAICompatibleProviderLabel,
+  listOpenAICompatibleModels,
+} from './providerDiscovery.js'
+
+const originalFetch = globalThis.fetch
+const originalEnv = {
+  OPENAI_BASE_URL: process.env.OPENAI_BASE_URL,
+}
+
+afterEach(() => {
+  globalThis.fetch = originalFetch
+  process.env.OPENAI_BASE_URL = originalEnv.OPENAI_BASE_URL
+})
+
+test('lists models from a local openai-compatible /models endpoint', async () => {
+  globalThis.fetch = mock((input, init) => {
+    const url = typeof input === 'string' ? input : input.url
+    expect(url).toBe('http://localhost:1234/v1/models')
+    expect(init?.headers).toEqual({ Authorization: 'Bearer local-key' })
+
+    return Promise.resolve(
+      new Response(
+        JSON.stringify({
+          data: [
+            { id: 'qwen2.5-coder-7b-instruct' },
+            { id: 'llama-3.2-3b-instruct' },
+            { id: 'qwen2.5-coder-7b-instruct' },
+          ],
+        }),
+        { status: 200 },
+      ),
+    )
+  }) as typeof globalThis.fetch
+
+  await expect(
+    listOpenAICompatibleModels({
+      baseUrl: 'http://localhost:1234/v1',
+      apiKey: 'local-key',
+    }),
+  ).resolves.toEqual([
+    'qwen2.5-coder-7b-instruct',
+    'llama-3.2-3b-instruct',
+  ])
+})
+
+test('returns null when a local openai-compatible /models request fails', async () => {
+  globalThis.fetch = mock(() =>
+    Promise.resolve(new Response('not available', { status: 503 })),
+  ) as typeof globalThis.fetch
+
+  await expect(
+    listOpenAICompatibleModels({ baseUrl: 'http://localhost:1234/v1' }),
+  ).resolves.toBeNull()
+})
+
+test('detects LM Studio from the default localhost port', () => {
+  expect(getLocalOpenAICompatibleProviderLabel('http://localhost:1234/v1')).toBe(
+    'LM Studio',
+  )
+})
+
+test('detects common local openai-compatible providers by hostname', () => {
+  expect(
+    getLocalOpenAICompatibleProviderLabel('http://localai.local:8080/v1'),
+  ).toBe('LocalAI')
+  expect(
+    getLocalOpenAICompatibleProviderLabel('http://vllm.local:8000/v1'),
+  ).toBe('vLLM')
+})
+
+test('falls back to a generic local openai-compatible label', () => {
+  expect(
+    getLocalOpenAICompatibleProviderLabel('http://127.0.0.1:8080/v1'),
+  ).toBe('Local OpenAI-compatible')
+})
--- a/src/utils/providerDiscovery.ts
+++ b/src/utils/providerDiscovery.ts
@@ -1,4 +1,5 @@
 import type { OllamaModelDescriptor } from './providerRecommendation.ts'
+import { DEFAULT_OPENAI_BASE_URL } from '../services/api/providerConfig.js'

 export const DEFAULT_OLLAMA_BASE_URL = 'http://localhost:11434'
 export const DEFAULT_ATOMIC_CHAT_BASE_URL = 'http://127.0.0.1:1337'
@@ -53,6 +54,64 @@ export function getAtomicChatChatBaseUrl(baseUrl?: string): string {
  return `${getAtomicChatApiBaseUrl(baseUrl)}/v1`
 }

+export function getOpenAICompatibleModelsBaseUrl(baseUrl?: string): string {
+  return (
+    baseUrl || process.env.OPENAI_BASE_URL || DEFAULT_OPENAI_BASE_URL
+  ).replace(/\/+$/, '')
+}
+
+export function getLocalOpenAICompatibleProviderLabel(baseUrl?: string): string {
+  try {
+    const parsed = new URL(getOpenAICompatibleModelsBaseUrl(baseUrl))
+    const host = parsed.host.toLowerCase()
+    const hostname = parsed.hostname.toLowerCase()
+    const path = parsed.pathname.toLowerCase()
+    const haystack = `${hostname} ${path}`
+
+    if (
+      host.endsWith(':1234') ||
+      haystack.includes('lmstudio') ||
+      haystack.includes('lm-studio')
+    ) {
+      return 'LM Studio'
+    }
+    if (host.endsWith(':11434') || haystack.includes('ollama')) {
+      return 'Ollama'
+    }
+    if (haystack.includes('localai')) {
+      return 'LocalAI'
+    }
+    if (haystack.includes('jan')) {
+      return 'Jan'
+    }
+    if (haystack.includes('kobold')) {
+      return 'KoboldCpp'
+    }
+    if (haystack.includes('llama.cpp') || haystack.includes('llamacpp')) {
+      return 'llama.cpp'
+    }
+    if (haystack.includes('vllm')) {
+      return 'vLLM'
+    }
+    if (
+      haystack.includes('open-webui') ||
+      haystack.includes('openwebui')
+    ) {
+      return 'Open WebUI'
+    }
+    if (
+      haystack.includes('text-generation-webui') ||
+      haystack.includes('oobabooga')
+    ) {
+      return 'text-generation-webui'
+    }
+  } catch {
+    // Fall back to the generic label when the base URL is malformed.
+  }
+
+  return 'Local OpenAI-compatible'
+}
+
 export async function hasLocalOllama(baseUrl?: string): Promise<boolean> {
  const { signal, clear } = withTimeoutSignal(1200)
  try {
@@ -111,6 +170,46 @@ export async function listOllamaModels(
  }
 }

+export async function listOpenAICompatibleModels(options?: {
+  baseUrl?: string
+  apiKey?: string
+}): Promise<string[] | null> {
+  const { signal, clear } = withTimeoutSignal(5000)
+  try {
+    const response = await fetch(
+      `${getOpenAICompatibleModelsBaseUrl(options?.baseUrl)}/models`,
+      {
+        method: 'GET',
+        headers: options?.apiKey
+          ? {
+              Authorization: `Bearer ${options.apiKey}`,
+            }
+          : undefined,
+        signal,
+      },
+    )
+    if (!response.ok) {
+      return null
+    }
+
+    const data = (await response.json()) as {
+      data?: Array<{ id?: string }>
+    }
+
+    return Array.from(
+      new Set(
+        (data.data ?? [])
+          .filter(model => Boolean(model.id))
+          .map(model => model.id!),
+      ),
+    )
+  } catch {
+    return null
+  } finally {
+    clear()
+  }
+}
+
 export async function hasLocalAtomicChat(baseUrl?: string): Promise<boolean> {
  const { signal, clear } = withTimeoutSignal(1200)
  try {