chore: release 0.1.8

Models served through Ollama/vLLM with strict Jinja templates (Devstral,
Mistral, etc.) require strict user↔assistant role alternation and reject
requests with consecutive messages of the same role.

convertMessages() could produce consecutive user or assistant messages in
three scenarios: batched user input, text-only + tool_use assistant turns,
and tool result remainders followed by another user message.

Added a coalescing pass at the end of convertMessages() that merges
consecutive same-role messages (string concat or array concat), preserving
tool_calls on assistant messages. Tool and system messages are excluded
from coalescing as they have their own alternation rules.

Includes regression tests for both user and assistant coalescing.

Fixes #202

bun.lock

@@ -51,7 +51,7 @@
         "ignore": "7.0.5",
         "indent-string": "5.0.0",
         "jsonc-parser": "3.3.1",
-        "lodash-es": "4.18.0",
+        "lodash-es": "4.18.1",
         "lru-cache": "11.2.7",
         "marked": "15.0.12",
         "p-map": "7.0.4",
@@ -88,6 +88,9 @@
       },
     },
   },
+  "overrides": {
+    "lodash-es": "4.18.1",
+  },
   "packages": {
     "@alcalzone/ansi-tokenize": ["@alcalzone/ansi-tokenize@0.3.0", "", { "dependencies": { "ansi-styles": "^6.2.1", "is-fullwidth-code-point": "^5.0.0" } }, "sha512-p+CMKJ93HFmLkjXKlXiVGlMQEuRb6H0MokBSwUsX+S6BRX8eV5naFZpQJFfJHjRZY0Hmnqy1/r6UWl3x+19zYA=="],
 
@@ -657,7 +660,7 @@
 
     "locate-path": ["locate-path@5.0.0", "", { "dependencies": { "p-locate": "^4.1.0" } }, "sha512-t7hw9pI+WvuwNJXwk5zVHpyhIqzg2qTlklJOf0mVxGSbe3Fp2VieZcduNYjaLDoy6p9uGpQEGWG87WpMKlNq8g=="],
 
-    "lodash-es": ["lodash-es@4.18.0", "", {}, "sha512-koAgswPPA+UTaPN64Etp+PGP+WT6oqOS2NMi5yDkMaiGw9qY4VxQbQF0mtKMyr4BlTznWyzePV5UpECTJQmSUA=="],
+    "lodash-es": ["lodash-es@4.18.1", "", {}, "sha512-J8xewKD/Gk22OZbhpOVSwcs60zhd95ESDwezOFuA3/099925PdHJ7OFHNTGtajL3AlZkykD32HykiMo+BIBI8A=="],
 
     "lodash.camelcase": ["lodash.camelcase@4.3.0", "", {}, "sha512-TwuEnCnxbc3rAvhf/LbG7tJUDzhqXyFnv3dtzLOPgCG/hODL7WFnsbwktkD7yUV0RrreP/l1PALq/YSg6VvjlA=="],
 
@@ -891,8 +894,6 @@
 
     "zod-to-json-schema": ["zod-to-json-schema@3.25.2", "", { "peerDependencies": { "zod": "^3.25.28 || ^4" } }, "sha512-O/PgfnpT1xKSDeQYSCfRI5Gy3hPf91mKVDuYLUHZJMiDFptvP41MSnWofm8dnCm0256ZNfZIM7DSzuSMAFnjHA=="],
 
-    "@anthropic-ai/sandbox-runtime/lodash-es": ["lodash-es@4.17.23", "", {}, "sha512-kVI48u3PZr38HdYz98UmfPnXl2DXrpdctLrFLCd3kOx1xUkOmpFPx7gCWWM5MPkL/fD8zb+Ph0QzjGFs4+hHWg=="],
-
     "@aws-crypto/crc32/@aws-crypto/util": ["@aws-crypto/util@5.2.0", "", { "dependencies": { "@aws-sdk/types": "^3.222.0", "@smithy/util-utf8": "^2.0.0", "tslib": "^2.6.2" } }, "sha512-4RkU9EsI6ZpBve5fseQlGNUWKMa1RLPQ1dnjnQoe07ldfIzcsGb5hC5W0Dm7u423KWzawlrpbjXBrXCEv9zazQ=="],
 
     "@aws-crypto/crc32/tslib": ["tslib@2.8.1", "", {}, "sha512-oJFu94HQb+KVduSUQL7wnpmqnfmLsOA/nAh6b6EH0wCEoK0/mPeXU6c3wKDV83MkOuHPRHtSXKKU99IBazS/2w=="],

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gitlawb/openclaude",
-  "version": "0.1.7",
+  "version": "0.1.8",
   "description": "Claude Code opened to any LLM — OpenAI, Gemini, DeepSeek, Ollama, and 200+ models",
   "type": "module",
   "bin": {
@@ -95,7 +95,7 @@
     "ignore": "7.0.5",
     "indent-string": "5.0.0",
     "jsonc-parser": "3.3.1",
-    "lodash-es": "4.18.0",
+    "lodash-es": "4.18.1",
     "lru-cache": "11.2.7",
     "marked": "15.0.12",
     "p-map": "7.0.4",
@@ -150,5 +150,8 @@
   "license": "SEE LICENSE FILE",
   "publishConfig": {
     "access": "public"
+  },
+  "overrides": {
+    "lodash-es": "4.18.1"
   }
 }

src/commands/model/model.test.tsx

@@ -0,0 +1,42 @@
+import { afterEach, expect, mock, test } from 'bun:test'
+
+const originalEnv = {
+  CLAUDE_CODE_USE_OPENAI: process.env.CLAUDE_CODE_USE_OPENAI,
+  OPENAI_BASE_URL: process.env.OPENAI_BASE_URL,
+  OPENAI_MODEL: process.env.OPENAI_MODEL,
+}
+
+afterEach(() => {
+  mock.restore()
+  process.env.CLAUDE_CODE_USE_OPENAI = originalEnv.CLAUDE_CODE_USE_OPENAI
+  process.env.OPENAI_BASE_URL = originalEnv.OPENAI_BASE_URL
+  process.env.OPENAI_MODEL = originalEnv.OPENAI_MODEL
+})
+
+test('opens the model picker without awaiting local model discovery refresh', async () => {
+  process.env.CLAUDE_CODE_USE_OPENAI = '1'
+  process.env.OPENAI_BASE_URL = 'http://127.0.0.1:8080/v1'
+  process.env.OPENAI_MODEL = 'qwen2.5-coder-7b-instruct'
+
+  let resolveDiscovery: (() => void) | undefined
+  const discoverOpenAICompatibleModelOptions = mock(
+    () =>
+      new Promise<void>(resolve => {
+        resolveDiscovery = resolve
+      }),
+  )
+
+  mock.module('../../utils/model/openaiModelDiscovery.js', () => ({
+    discoverOpenAICompatibleModelOptions,
+  }))
+
+  const { call } = await import(`./model.js?ts=${Date.now()}-${Math.random()}`)
+  const result = await Promise.race([
+    call(() => {}, {} as never, ''),
+    new Promise(resolve => setTimeout(() => resolve('timeout'), 50)),
+  ])
+
+  resolveDiscovery?.()
+
+  expect(result).not.toBe('timeout')
+})

src/commands/model/model.tsx

@@ -4,6 +4,7 @@ import * as React from 'react';
 import type { CommandResultDisplay } from '../../commands.js';
 import { ModelPicker } from '../../components/ModelPicker.js';
 import { COMMON_HELP_ARGS, COMMON_INFO_ARGS } from '../../constants/xml.js';
+import { fetchBootstrapData } from '../../services/api/bootstrap.js';
 import { type AnalyticsMetadata_I_VERIFIED_THIS_IS_NOT_CODE_OR_FILEPATHS, logEvent } from '../../services/analytics/index.js';
 import { useAppState, useSetAppState } from '../../state/AppState.js';
 import type { LocalJSXCommandCall } from '../../types/command.js';
@@ -19,6 +20,7 @@ import { getActiveOpenAIModelOptionsCache, setActiveOpenAIModelOptionsCache } fr
 import { getDefaultMainLoopModelSetting, isOpus1mMergeEnabled, renderDefaultModelSetting } from '../../utils/model/model.js';
 import { isModelAllowed } from '../../utils/model/modelAllowlist.js';
 import { validateModel } from '../../utils/model/validateModel.js';
+import { getAdditionalModelOptionsCacheScope } from '../../services/api/providerConfig.js';
 function ModelPickerWrapper(t0) {
   const $ = _c(17);
   const {
@@ -319,7 +321,9 @@ export const call: LocalJSXCommandCall = async (onDone, _context, args) => {
     });
     return <SetModelAndClose args={args} onDone={onDone} />;
   }
-  await refreshOpenAIModelOptionsCache();
+  if (getAdditionalModelOptionsCacheScope()?.startsWith('openai:')) {
+    void refreshOpenAIModelOptionsCache();
+  }
   return <ModelPickerWrapper onDone={onDone} />;
 };
 function renderModelLabel(model: string | null): string {

src/commands/provider/provider.test.tsx

@@ -197,6 +197,21 @@ test('buildProfileSaveMessage maps provider fields without echoing secrets', ()
   expect(message).not.toContain('sk-secret-12345678')
 })
 
+test('buildProfileSaveMessage labels local openai-compatible profiles consistently', () => {
+  const message = buildProfileSaveMessage(
+    'openai',
+    {
+      OPENAI_MODEL: 'gpt-5.4',
+      OPENAI_BASE_URL: 'http://127.0.0.1:8080/v1',
+    },
+    'D:/codings/Opensource/openclaude/.openclaude-profile.json',
+  )
+
+  expect(message).toContain('Saved Local OpenAI-compatible profile.')
+  expect(message).toContain('Model: gpt-5.4')
+  expect(message).toContain('Endpoint: http://127.0.0.1:8080/v1')
+})
+
 test('buildProfileSaveMessage describes Gemini access token / ADC mode clearly', () => {
   const message = buildProfileSaveMessage(
     'gemini',
@@ -230,6 +245,36 @@ test('buildCurrentProviderSummary redacts poisoned model and endpoint values', (
   expect(summary.endpointLabel).toBe('sk-...5678')
 })
 
+test('buildCurrentProviderSummary labels generic local openai-compatible providers', () => {
+  const summary = buildCurrentProviderSummary({
+    processEnv: {
+      CLAUDE_CODE_USE_OPENAI: '1',
+      OPENAI_MODEL: 'qwen2.5-coder-7b-instruct',
+      OPENAI_BASE_URL: 'http://127.0.0.1:8080/v1',
+    },
+    persisted: null,
+  })
+
+  expect(summary.providerLabel).toBe('Local OpenAI-compatible')
+  expect(summary.modelLabel).toBe('qwen2.5-coder-7b-instruct')
+  expect(summary.endpointLabel).toBe('http://127.0.0.1:8080/v1')
+})
+
+test('buildCurrentProviderSummary does not relabel local gpt-5.4 providers as Codex', () => {
+  const summary = buildCurrentProviderSummary({
+    processEnv: {
+      CLAUDE_CODE_USE_OPENAI: '1',
+      OPENAI_MODEL: 'gpt-5.4',
+      OPENAI_BASE_URL: 'http://127.0.0.1:8080/v1',
+    },
+    persisted: null,
+  })
+
+  expect(summary.providerLabel).toBe('Local OpenAI-compatible')
+  expect(summary.modelLabel).toBe('gpt-5.4')
+  expect(summary.endpointLabel).toBe('http://127.0.0.1:8080/v1')
+})
+
 test('getProviderWizardDefaults ignores poisoned current provider values', () => {
   const defaults = getProviderWizardDefaults({
     OPENAI_API_KEY: 'sk-secret-12345678',

src/commands/provider/provider.tsx

@@ -15,6 +15,7 @@ import { Box, Text } from '../../ink.js'
 import {
   DEFAULT_CODEX_BASE_URL,
   DEFAULT_OPENAI_BASE_URL,
+  isLocalProviderUrl,
   resolveCodexApiCredentials,
   resolveProviderRequest,
 } from '../../services/api/providerConfig.js'
@@ -52,7 +53,11 @@ import {
   recommendOllamaModel,
   type RecommendationGoal,
 } from '../../utils/providerRecommendation.js'
-import { hasLocalOllama, listOllamaModels } from '../../utils/providerDiscovery.js'
+import {
+  getLocalOpenAICompatibleProviderLabel,
+  hasLocalOllama,
+  listOllamaModels,
+} from '../../utils/providerDiscovery.js'
 
 type ProviderChoice = 'auto' | ProviderProfile | 'clear'
 
@@ -182,10 +187,8 @@ export function buildCurrentProviderSummary(options?: {
     let providerLabel = 'OpenAI-compatible'
     if (request.transport === 'codex_responses') {
       providerLabel = 'Codex'
-    } else if (request.baseUrl.includes('localhost:11434')) {
-      providerLabel = 'Ollama'
-    } else if (request.baseUrl.includes('localhost:1234')) {
-      providerLabel = 'LM Studio'
+    } else if (isLocalProviderUrl(request.baseUrl)) {
+      providerLabel = getLocalOpenAICompatibleProviderLabel(request.baseUrl)
     }
 
     return {
@@ -272,16 +275,20 @@ function buildSavedProfileSummary(
         ),
       }
     case 'openai':
-    default:
+    default: {
+      const baseUrl = env.OPENAI_BASE_URL ?? DEFAULT_OPENAI_BASE_URL
+
       return {
-        providerLabel: 'OpenAI-compatible',
+        providerLabel: isLocalProviderUrl(baseUrl)
+          ? getLocalOpenAICompatibleProviderLabel(baseUrl)
+          : 'OpenAI-compatible',
         modelLabel: getSafeDisplayValue(
           env.OPENAI_MODEL ?? 'gpt-4o',
           process.env,
           env,
         ),
         endpointLabel: getSafeDisplayValue(
-          env.OPENAI_BASE_URL ?? DEFAULT_OPENAI_BASE_URL,
+          baseUrl,
           process.env,
           env,
         ),
@@ -292,6 +299,7 @@ function buildSavedProfileSummary(
       }
     }
   }
+}
 
 export function buildProfileSaveMessage(
   profile: ProviderProfile,

src/components/StartupScreen.ts

@@ -5,6 +5,9 @@
  * Addresses: https://github.com/Gitlawb/openclaude/issues/55
  */
 
+import { isLocalProviderUrl } from '../services/api/providerConfig.js'
+import { getLocalOpenAICompatibleProviderLabel } from '../utils/providerDiscovery.js'
+
 declare const MACRO: { VERSION: string; DISPLAY_VERSION?: string }
 
 const ESC = '\x1b['
@@ -99,7 +102,7 @@ function detectProvider(): { name: string; model: string; baseUrl: string; isLoc
   if (useOpenAI) {
     const rawModel = process.env.OPENAI_MODEL || 'gpt-4o'
     const baseUrl = process.env.OPENAI_BASE_URL || 'https://api.openai.com/v1'
-    const isLocal = /localhost|127\.0\.0\.1|0\.0\.0\.0/.test(baseUrl)
+    const isLocal = isLocalProviderUrl(baseUrl)
     let name = 'OpenAI'
     if (/deepseek/i.test(baseUrl) || /deepseek/i.test(rawModel))       name = 'DeepSeek'
     else if (/openrouter/i.test(baseUrl))                             name = 'OpenRouter'
@@ -107,10 +110,8 @@ function detectProvider(): { name: string; model: string; baseUrl: string; isLoc
     else if (/groq/i.test(baseUrl))                                   name = 'Groq'
     else if (/mistral/i.test(baseUrl) || /mistral/i.test(rawModel))     name = 'Mistral'
     else if (/azure/i.test(baseUrl))                                  name = 'Azure OpenAI'
-    else if (/localhost:11434/i.test(baseUrl))                        name = 'Ollama'
-    else if (/localhost:1234/i.test(baseUrl))                         name = 'LM Studio'
     else if (/llama/i.test(rawModel))                                    name = 'Meta Llama'
-    else if (isLocal)                                                  name = 'Local'
+    else if (isLocal)                                                  name = getLocalOpenAICompatibleProviderLabel(baseUrl)
     
     // Resolve model alias to actual model name + reasoning effort
     let displayModel = rawModel

src/services/api/bootstrap.ts

@@ -14,7 +14,16 @@ import { lazySchema } from '../../utils/lazySchema.js'
 import { logError } from '../../utils/log.js'
 import { getAPIProvider } from '../../utils/model/providers.js'
 import { isEssentialTrafficOnly } from '../../utils/privacyLevel.js'
+import type { ModelOption } from '../../utils/model/modelOptions.js'
+import {
+  getLocalOpenAICompatibleProviderLabel,
+  listOpenAICompatibleModels,
+} from '../../utils/providerDiscovery.js'
 import { getClaudeCodeUserAgent } from '../../utils/userAgent.js'
+import {
+  getAdditionalModelOptionsCacheScope,
+  resolveProviderRequest,
+} from './providerConfig.js'
 
 const bootstrapResponseSchema = lazySchema(() =>
   z.object({
@@ -39,6 +48,12 @@ const bootstrapResponseSchema = lazySchema(() =>
 
 type BootstrapResponse = z.infer<ReturnType<typeof bootstrapResponseSchema>>
 
+type BootstrapCachePayload = {
+  clientData: Record<string, unknown> | null
+  additionalModelOptions: ModelOption[]
+  additionalModelOptionsScope: string
+}
+
 async function fetchBootstrapAPI(): Promise<BootstrapResponse | null> {
   if (isEssentialTrafficOnly()) {
     logForDebugging('[Bootstrap] Skipped: Nonessential traffic disabled')
@@ -108,22 +123,70 @@ async function fetchBootstrapAPI(): Promise<BootstrapResponse | null> {
   }
 }
 
+async function fetchLocalOpenAIModelOptions(): Promise<BootstrapCachePayload | null> {
+  const scope = getAdditionalModelOptionsCacheScope()
+  if (!scope?.startsWith('openai:')) {
+    return null
+  }
+
+  const { baseUrl } = resolveProviderRequest()
+  const models = await listOpenAICompatibleModels({
+    baseUrl,
+    apiKey: process.env.OPENAI_API_KEY,
+  })
+
+  if (models === null) {
+    logForDebugging('[Bootstrap] Local OpenAI model discovery failed')
+    return null
+  }
+
+  const providerLabel = getLocalOpenAICompatibleProviderLabel(baseUrl)
+
+  return {
+    clientData: getGlobalConfig().clientDataCache ?? null,
+    additionalModelOptionsScope: scope,
+    additionalModelOptions: models.map(model => ({
+      value: model,
+      label: model,
+      description: `Detected from ${providerLabel}`,
+    })),
+  }
+}
+
 /**
  * Fetch bootstrap data from the API and persist to disk cache.
  */
 export async function fetchBootstrapData(): Promise<void> {
   try {
+    const scope = getAdditionalModelOptionsCacheScope()
+    let payload: BootstrapCachePayload | null = null
+
+    if (scope === 'firstParty') {
       const response = await fetchBootstrapAPI()
       if (!response) return
 
-    const clientData = response.client_data ?? null
-    const additionalModelOptions = response.additional_model_options ?? []
+      payload = {
+        clientData: response.client_data ?? null,
+        additionalModelOptions: response.additional_model_options ?? [],
+        additionalModelOptionsScope: scope,
+      }
+    } else if (scope?.startsWith('openai:')) {
+      payload = await fetchLocalOpenAIModelOptions()
+      if (!payload) return
+    } else {
+      logForDebugging('[Bootstrap] Skipped: no additional model source')
+      return
+    }
+
+    const { clientData, additionalModelOptions, additionalModelOptionsScope } =
+      payload
 
     // Only persist if data actually changed — avoids a config write on every startup.
     const config = getGlobalConfig()
     if (
       isEqual(config.clientDataCache, clientData) &&
-      isEqual(config.additionalModelOptionsCache, additionalModelOptions)
+      isEqual(config.additionalModelOptionsCache, additionalModelOptions) &&
+      config.additionalModelOptionsCacheScope === additionalModelOptionsScope
     ) {
       logForDebugging('[Bootstrap] Cache unchanged, skipping write')
       return
@@ -134,6 +197,7 @@ export async function fetchBootstrapData(): Promise<void> {
       ...current,
       clientDataCache: clientData,
       additionalModelOptionsCache: additionalModelOptions,
+      additionalModelOptionsCacheScope: additionalModelOptionsScope,
     }))
   } catch (error) {
     logError(error)

src/services/api/codexShim.test.ts

@@ -14,12 +14,19 @@ import {
 } from './providerConfig.js'
 
 const tempDirs: string[] = []
+const originalEnv = {
+  OPENAI_BASE_URL: process.env.OPENAI_BASE_URL,
+  OPENAI_API_BASE: process.env.OPENAI_API_BASE,
+}
 
 afterEach(() => {
   while (tempDirs.length > 0) {
     const dir = tempDirs.pop()
     if (dir) rmSync(dir, { recursive: true, force: true })
   }
+
+  process.env.OPENAI_BASE_URL = originalEnv.OPENAI_BASE_URL
+  process.env.OPENAI_API_BASE = originalEnv.OPENAI_API_BASE
 })
 
 function createTempAuthJson(payload: Record<string, unknown>): string {
@@ -62,12 +69,26 @@ describe('Codex provider config', () => {
   })
 
   test('resolves codexplan alias to Codex transport with reasoning', () => {
+    delete process.env.OPENAI_BASE_URL
+    delete process.env.OPENAI_API_BASE
+
     const resolved = resolveProviderRequest({ model: 'codexplan' })
     expect(resolved.transport).toBe('codex_responses')
     expect(resolved.resolvedModel).toBe('gpt-5.4')
     expect(resolved.reasoning).toEqual({ effort: 'high' })
   })
 
+  test('does not force Codex transport when a local non-Codex base URL is explicit', () => {
+    const resolved = resolveProviderRequest({
+      model: 'codexplan',
+      baseUrl: 'http://127.0.0.1:8080/v1',
+    })
+
+    expect(resolved.transport).toBe('chat_completions')
+    expect(resolved.baseUrl).toBe('http://127.0.0.1:8080/v1')
+    expect(resolved.resolvedModel).toBe('gpt-5.4')
+  })
+
   test('resolves codexplan to Codex transport even when OPENAI_BASE_URL is the string "undefined"', () => {
     // On Windows, env vars can leak as the literal string "undefined" instead of
     // the JS value undefined when not properly unset (issue #336).

src/services/api/errors.ts

@@ -557,8 +557,12 @@ export function getAssistantMessageFromError(
     const stripped = error.message.replace(/^429\s+/, '')
     const innerMessage = stripped.match(/"message"\s*:\s*"([^"]*)"/)?.[1]
     const detail = innerMessage || stripped
+    const retryAfter = (error as APIError).headers?.get?.('retry-after')
+    const retryHint = retryAfter && !isNaN(Number(retryAfter))
+      ? `Try again in ${retryAfter} seconds.`
+      : 'Try again in a few seconds.'
     return createAssistantAPIErrorMessage({
-      content: `${API_ERROR_MESSAGE_PREFIX}: Request rejected (429) · ${detail || `this may be a temporary capacity issue${getAPIProvider() === 'firstParty' ? ' — check status.anthropic.com' : ''}`}`,
+      content: `${API_ERROR_MESSAGE_PREFIX}: Request rejected (429) · ${detail || 'this may be a temporary capacity issue'} — ${retryHint}`,
       error: 'rate_limit',
     })
   }

src/services/api/openaiShim.test.ts

@@ -573,3 +573,80 @@ test('sanitizes malformed MCP tool schemas before sending them to OpenAI', async
   expect(properties?.priority?.enum).toEqual([0, 1, 2, 3])
   expect(properties?.priority).not.toHaveProperty('default')
 })
+
+// ---------------------------------------------------------------------------
+// Issue #202 — consecutive role coalescing (Devstral, Mistral strict templates)
+// ---------------------------------------------------------------------------
+
+function makeNonStreamResponse(content = 'ok'): Response {
+  return new Response(
+    JSON.stringify({
+      id: 'chatcmpl-test',
+      model: 'test-model',
+      choices: [{ message: { role: 'assistant', content }, finish_reason: 'stop' }],
+      usage: { prompt_tokens: 5, completion_tokens: 1, total_tokens: 6 },
+    }),
+    { headers: { 'Content-Type': 'application/json' } },
+  )
+}
+
+test('coalesces consecutive user messages to avoid alternation errors (issue #202)', async () => {
+  let sentMessages: Array<{ role: string; content: unknown }> | undefined
+
+  globalThis.fetch = (async (_input: unknown, init: RequestInit | undefined) => {
+    sentMessages = JSON.parse(String(init?.body)).messages
+    return makeNonStreamResponse()
+  }) as FetchType
+
+  const client = createOpenAIShimClient({}) as OpenAIShimClient
+
+  await client.beta.messages.create({
+    model: 'test-model',
+    system: 'sys',
+    messages: [
+      { role: 'user', content: 'first message' },
+      { role: 'user', content: 'second message' },
+    ],
+    max_tokens: 64,
+    stream: false,
+  })
+
+  expect(sentMessages?.length).toBe(2) // system + 1 merged user
+  expect(sentMessages?.[0]?.role).toBe('system')
+  expect(sentMessages?.[1]?.role).toBe('user')
+  const userContent = sentMessages?.[1]?.content as string
+  expect(userContent).toContain('first message')
+  expect(userContent).toContain('second message')
+})
+
+test('coalesces consecutive assistant messages preserving tool_calls (issue #202)', async () => {
+  let sentMessages: Array<{ role: string; content: unknown; tool_calls?: unknown[] }> | undefined
+
+  globalThis.fetch = (async (_input: unknown, init: RequestInit | undefined) => {
+    sentMessages = JSON.parse(String(init?.body)).messages
+    return makeNonStreamResponse()
+  }) as FetchType
+
+  const client = createOpenAIShimClient({}) as OpenAIShimClient
+
+  await client.beta.messages.create({
+    model: 'test-model',
+    system: 'sys',
+    messages: [
+      { role: 'user', content: 'go' },
+      { role: 'assistant', content: 'thinking...' },
+      {
+        role: 'assistant',
+        content: [{ type: 'tool_use', id: 'call_1', name: 'Bash', input: { command: 'ls' } }],
+      },
+      { role: 'user', content: [{ type: 'tool_result', tool_use_id: 'call_1', content: 'file.txt' }] },
+    ],
+    max_tokens: 64,
+    stream: false,
+  })
+
+  // system + user + merged assistant + tool
+  const assistantMsgs = sentMessages?.filter(m => m.role === 'assistant')
+  expect(assistantMsgs?.length).toBe(1) // two assistant turns merged into one
+  expect(assistantMsgs?.[0]?.tool_calls?.length).toBeGreaterThan(0)
+})

src/services/api/openaiShim.ts

@@ -295,7 +295,41 @@ function convertMessages(
     }
   }
 
-  return result
+  // Coalescing pass: merge consecutive messages of the same role.
+  // OpenAI/vLLM/Ollama require strict user↔assistant alternation.
+  // Multiple consecutive tool messages are allowed (assistant → tool* → user).
+  // Consecutive user or assistant messages must be merged to avoid Jinja
+  // template errors like "roles must alternate" (Devstral, Mistral models).
+  const coalesced: OpenAIMessage[] = []
+  for (const msg of result) {
+    const prev = coalesced[coalesced.length - 1]
+
+    if (prev && prev.role === msg.role && msg.role !== 'tool' && msg.role !== 'system') {
+      const prevContent = prev.content
+      const curContent = msg.content
+
+      if (typeof prevContent === 'string' && typeof curContent === 'string') {
+        prev.content = prevContent + (prevContent && curContent ? '\n' : '') + curContent
+      } else {
+        const toArray = (
+          c: string | Array<{ type: string; text?: string; image_url?: { url: string } }> | undefined,
+        ): Array<{ type: string; text?: string; image_url?: { url: string } }> => {
+          if (!c) return []
+          if (typeof c === 'string') return c ? [{ type: 'text', text: c }] : []
+          return c
+        }
+        prev.content = [...toArray(prevContent), ...toArray(curContent)]
+      }
+
+      if (msg.tool_calls?.length) {
+        prev.tool_calls = [...(prev.tool_calls ?? []), ...msg.tool_calls]
+      }
+    } else {
+      coalesced.push(msg)
+    }
+  }
+
+  return coalesced
 }
 
 /**

src/services/api/providerConfig.local.test.ts

@@ -1,6 +1,22 @@
-import { expect, test } from 'bun:test'
+import { afterEach, expect, test } from 'bun:test'
 
-import { isLocalProviderUrl } from './providerConfig.js'
+import {
+  getAdditionalModelOptionsCacheScope,
+  isLocalProviderUrl,
+  resolveProviderRequest,
+} from './providerConfig.js'
+
+const originalEnv = {
+  CLAUDE_CODE_USE_OPENAI: process.env.CLAUDE_CODE_USE_OPENAI,
+  OPENAI_BASE_URL: process.env.OPENAI_BASE_URL,
+  OPENAI_MODEL: process.env.OPENAI_MODEL,
+}
+
+afterEach(() => {
+  process.env.CLAUDE_CODE_USE_OPENAI = originalEnv.CLAUDE_CODE_USE_OPENAI
+  process.env.OPENAI_BASE_URL = originalEnv.OPENAI_BASE_URL
+  process.env.OPENAI_MODEL = originalEnv.OPENAI_MODEL
+})
 
 test('treats localhost endpoints as local', () => {
   expect(isLocalProviderUrl('http://localhost:11434/v1')).toBe(true)
@@ -33,3 +49,37 @@ test('treats public hosts as remote', () => {
   expect(isLocalProviderUrl('https://example.com/v1')).toBe(false)
   expect(isLocalProviderUrl('http://[2001:4860:4860::8888]:11434/v1')).toBe(false)
 })
+
+test('creates a cache scope for local openai-compatible providers', () => {
+  process.env.CLAUDE_CODE_USE_OPENAI = '1'
+  process.env.OPENAI_BASE_URL = 'http://localhost:1234/v1'
+  process.env.OPENAI_MODEL = 'llama-3.2-3b-instruct'
+
+  expect(getAdditionalModelOptionsCacheScope()).toBe(
+    'openai:http://localhost:1234/v1',
+  )
+})
+
+test('keeps codex alias models on chat completions for local openai-compatible providers', () => {
+  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(resolveProviderRequest()).toMatchObject({
+    transport: 'chat_completions',
+    requestedModel: 'gpt-5.4',
+    resolvedModel: 'gpt-5.4',
+    baseUrl: 'http://127.0.0.1:8080/v1',
+  })
+  expect(getAdditionalModelOptionsCacheScope()).toBe(
+    'openai:http://127.0.0.1:8080/v1',
+  )
+})
+
+test('skips local model cache scope for remote openai-compatible providers', () => {
+  process.env.CLAUDE_CODE_USE_OPENAI = '1'
+  process.env.OPENAI_BASE_URL = 'https://api.openai.com/v1'
+  process.env.OPENAI_MODEL = 'gpt-4o'
+
+  expect(getAdditionalModelOptionsCacheScope()).toBeNull()
+})

src/services/api/providerConfig.ts

@@ -219,6 +219,14 @@ export function isCodexAlias(model: string): boolean {
   return base in CODEX_ALIAS_MODELS
 }
 
+export function shouldUseCodexTransport(
+  model: string,
+  baseUrl: string | undefined,
+): boolean {
+  const explicitBaseUrl = asEnvUrl(baseUrl)
+  return isCodexBaseUrl(explicitBaseUrl) || (!explicitBaseUrl && isCodexAlias(model))
+}
+
 export function isLocalProviderUrl(baseUrl: string | undefined): boolean {
   if (!baseUrl) return false
   try {
@@ -302,13 +310,8 @@ export function resolveProviderRequest(options?: {
     asEnvUrl(options?.baseUrl) ??
     asEnvUrl(process.env.OPENAI_BASE_URL) ??
     asEnvUrl(process.env.OPENAI_API_BASE)
-  // Use Codex transport only when:
-  // - the base URL is explicitly the Codex endpoint, OR
-  // - the model is a Codex alias AND no custom base URL has been set
-  // A custom OPENAI_BASE_URL (e.g. Azure, OpenRouter) always wins over
-  // model-name-based Codex detection to prevent auth failures (#200, #203).
   const transport: ProviderTransport =
-    isCodexBaseUrl(rawBaseUrl) || (!rawBaseUrl && isCodexAlias(requestedModel))
+    shouldUseCodexTransport(requestedModel, rawBaseUrl)
       ? 'codex_responses'
       : 'chat_completions'
 
@@ -337,6 +340,30 @@ export function resolveProviderRequest(options?: {
   }
 }
 
+export function getAdditionalModelOptionsCacheScope(): string | null {
+  if (!isEnvTruthy(process.env.CLAUDE_CODE_USE_OPENAI)) {
+    if (!isEnvTruthy(process.env.CLAUDE_CODE_USE_GEMINI) &&
+        !isEnvTruthy(process.env.CLAUDE_CODE_USE_GITHUB) &&
+        !isEnvTruthy(process.env.CLAUDE_CODE_USE_BEDROCK) &&
+        !isEnvTruthy(process.env.CLAUDE_CODE_USE_VERTEX) &&
+        !isEnvTruthy(process.env.CLAUDE_CODE_USE_FOUNDRY)) {
+      return 'firstParty'
+    }
+    return null
+  }
+
+  const request = resolveProviderRequest()
+  if (request.transport !== 'chat_completions') {
+    return null
+  }
+
+  if (!isLocalProviderUrl(request.baseUrl)) {
+    return null
+  }
+
+  return `openai:${request.baseUrl.toLowerCase()}`
+}
+
 export function resolveCodexAuthPath(
   env: NodeJS.ProcessEnv = process.env,
 ): string {

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[]

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)
     }

src/utils/model/openaiContextWindows.ts

@@ -23,9 +23,13 @@ const OPENAI_CONTEXT_WINDOWS: Record<string, number> = {
   'gpt-4.1-nano':             1_047_576,
   'gpt-4-turbo':              128_000,
   'gpt-4':                     8_192,
+  'o1':                       200_000,
+  'o1-mini':                  128_000,
+  'o1-preview':               128_000,
+  'o1-pro':                   200_000,
+  'o3':                       200_000,
   'o3-mini':                  200_000,
   'o4-mini':                  200_000,
-  'o3':                       200_000,
 
   // DeepSeek (V3: 128k context per official docs)
   'deepseek-chat':            128_000,
@@ -63,6 +67,9 @@ const OPENAI_CONTEXT_WINDOWS: Record<string, number> = {
   'phi4:14b':                  16_384,
   'gemma2:27b':                 8_192,
   'codellama:13b':              16_384,
+  'llama3.2:1b':              128_000,
+  'qwen3:8b':                 128_000,
+  'codestral':                 32_768,
 }
 
 /**
@@ -82,9 +89,13 @@ const OPENAI_MAX_OUTPUT_TOKENS: Record<string, number> = {
   'gpt-4.1-nano':             32_768,
   'gpt-4-turbo':               4_096,
   'gpt-4':                     4_096,
+  'o1':                       100_000,
+  'o1-mini':                   65_536,
+  'o1-preview':                32_768,
+  'o1-pro':                   100_000,
+  'o3':                       100_000,
   'o3-mini':                  100_000,
   'o4-mini':                  100_000,
-  'o3':                       100_000,
 
   // DeepSeek
   'deepseek-chat':              8_192,
@@ -120,6 +131,9 @@ const OPENAI_MAX_OUTPUT_TOKENS: Record<string, number> = {
   'phi4:14b':                   4_096,
   'gemma2:27b':                 4_096,
   'codellama:13b':              4_096,
+  'llama3.2:1b':                4_096,
+  'qwen3:8b':                   8_192,
+  'codestral':                   8_192,
 }
 
 function lookupByModel<T>(table: Record<string, T>, model: string): T | undefined {

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 +78,32 @@ 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', async () => {
+  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'
+
+  const { getAPIProvider } = await importFreshProvidersModule()
+  expect(getAPIProvider()).toBe('openai')
+})
+
+test('codex aliases still resolve to the codex provider without a non-codex base URL', async () => {
+  clearProviderEnv()
+  process.env.CLAUDE_CODE_USE_OPENAI = '1'
+  process.env.OPENAI_MODEL = 'codexplan'
+
+  const { getAPIProvider } = await importFreshProvidersModule()
+  expect(getAPIProvider()).toBe('codex')
+})
+
+test('official OpenAI base URLs now keep provider detection on openai for aliases', async () => {
+  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'
+
+  const { getAPIProvider } = await importFreshProvidersModule()
+  expect(getAPIProvider()).toBe('openai')
+})

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 {

src/utils/plugins/pluginLoader.test.ts

@@ -0,0 +1,71 @@
+import { describe, expect, test } from 'bun:test'
+
+import type { LoadedPlugin } from '../../types/plugin.js'
+import { mergePluginSources } from './pluginLoader.js'
+
+function marketplacePlugin(
+  name: string,
+  marketplace: string,
+  enabled: boolean,
+): LoadedPlugin {
+  const pluginId = `${name}@${marketplace}`
+  return {
+    name,
+    manifest: { name } as LoadedPlugin['manifest'],
+    path: `/tmp/${pluginId}`,
+    source: pluginId,
+    repository: pluginId,
+    enabled,
+  }
+}
+
+describe('mergePluginSources', () => {
+  test('keeps the enabled copy when duplicate marketplace plugins disagree on enabled state', () => {
+    const enabledOfficial = marketplacePlugin(
+      'frontend-design',
+      'claude-plugins-official',
+      true,
+    )
+    const disabledLegacy = marketplacePlugin(
+      'frontend-design',
+      'claude-code-plugins',
+      false,
+    )
+
+    const result = mergePluginSources({
+      session: [],
+      marketplace: [disabledLegacy, enabledOfficial],
+      builtin: [],
+    })
+
+    expect(result.plugins).toEqual([enabledOfficial])
+    expect(result.errors).toEqual([])
+  })
+
+  test('keeps the later copy when duplicate marketplace plugins are both enabled', () => {
+    const legacy = marketplacePlugin(
+      'frontend-design',
+      'claude-code-plugins',
+      true,
+    )
+    const official = marketplacePlugin(
+      'frontend-design',
+      'claude-plugins-official',
+      true,
+    )
+
+    const result = mergePluginSources({
+      session: [],
+      marketplace: [legacy, official],
+      builtin: [],
+    })
+
+    expect(result.plugins).toEqual([official])
+    expect(result.errors).toHaveLength(1)
+    expect(result.errors[0]).toMatchObject({
+      type: 'generic-error',
+      source: legacy.source,
+      plugin: legacy.name,
+    })
+  })
+})

src/utils/plugins/pluginLoader.ts

@@ -3045,24 +3045,63 @@ export function mergePluginSources(sources: {
   })
 
   const sessionNames = new Set(sessionPlugins.map(p => p.name))
-  const marketplacePlugins = sources.marketplace.filter(p => {
-    if (sessionNames.has(p.name)) {
+  // Different marketplaces can enable the same short plugin name, but
+  // downstream command/skill loading scopes by plugin.name.
+  const marketplacePluginsByName = new Map<string, LoadedPlugin>()
+  for (const plugin of sources.marketplace) {
+    if (sessionNames.has(plugin.name)) {
       logForDebugging(
-        `Plugin "${p.name}" from --plugin-dir overrides installed version`,
+        `Plugin "${plugin.name}" from --plugin-dir overrides installed version`,
       )
-      return false
+      continue
     }
-    return true
+    const existing = marketplacePluginsByName.get(plugin.name)
+    if (!existing) {
+      marketplacePluginsByName.set(plugin.name, plugin)
+      continue
+    }
+
+    const winner = selectMarketplacePlugin(existing, plugin)
+    const dropped = winner === existing ? plugin : existing
+    marketplacePluginsByName.set(plugin.name, winner)
+
+    logForDebugging(
+      `Ignoring duplicate marketplace plugin "${plugin.name}" from ${dropped.source}; using ${winner.source}`,
+      { level: 'warn' },
+    )
+    if (existing.enabled && plugin.enabled) {
+      errors.push({
+        type: 'generic-error',
+        source: dropped.source,
+        plugin: plugin.name,
+        error: `Duplicate marketplace plugin "${plugin.name}" ignored: using "${winner.source}" and skipping "${dropped.source}" to avoid short-name collisions`,
       })
+    }
+  }
   // Session first, then non-overridden marketplace, then builtin.
   // Downstream first-match consumers see session plugins before
   // installed ones for any that slipped past the name filter.
   return {
-    plugins: [...sessionPlugins, ...marketplacePlugins, ...sources.builtin],
+    plugins: [
+      ...sessionPlugins,
+      ...marketplacePluginsByName.values(),
+      ...sources.builtin,
+    ],
     errors,
   }
 }
 
+function selectMarketplacePlugin(
+  current: LoadedPlugin,
+  candidate: LoadedPlugin,
+): LoadedPlugin {
+  if (current.enabled !== candidate.enabled) {
+    return candidate.enabled ? candidate : current
+  }
+
+  return candidate
+}
+
 /**
  * Main plugin loading function that discovers and loads all plugins.
  *

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')
+})

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 {