fix(coding-agent): persist API mode per session

docs/openapi.json

@@ -9267,6 +9267,22 @@
                   },
                   "provider": {
                     "type": "string"
+                  },
+                  "apiMode": {
+                    "type": "string",
+                    "enum": [
+                      "chat_completions",
+                      "codex_responses",
+                      "anthropic_messages"
+                    ]
+                  },
+                  "api_mode": {
+                    "type": "string",
+                    "enum": [
+                      "chat_completions",
+                      "codex_responses",
+                      "anthropic_messages"
+                    ]
                   }
                 },
                 "required": [
@@ -11690,4 +11706,4 @@
       }
     }
   }
-}
+}

packages/client/src/api/coding-agents.ts

@@ -106,6 +106,7 @@ export interface CodingAgentLaunchResult {
   profile: string
   provider: string
   model: string
+  apiMode?: CodingAgentApiMode
   rootDir: string
   workspaceDir: string
   command: string

packages/client/src/api/hermes/sessions.ts

@@ -11,6 +11,7 @@ export interface SessionSummary {
   agent_native_session_id?: string
   model: string
   provider?: string
+  api_mode?: ProviderApiMode
   title: string | null
   parent_session_id?: string | null
   fork_point_message_id?: string | null

packages/client/src/components/hermes/chat/ChatPanel.vue

@@ -1078,7 +1078,9 @@ async function selectSessionModel(model: string, provider: string) {
   if (meta?.disabled || !sessionModelSessionId.value) return;
   if (isSessionModelExternalCodingAgent.value) {
     pendingSessionModelSwitch.value = { model, provider };
-    sessionModelApiMode.value = defaultSessionModelApiMode(provider);
+    sessionModelApiMode.value = sessionModelSession.value?.provider === provider && sessionModelSession.value.apiMode
+      ? normalizeCodingAgentApiMode(sessionModelSession.value.apiMode, defaultSessionModelApiMode(provider))
+      : defaultSessionModelApiMode(provider);
     showSessionModelModeModal.value = true;
     return;
   }

packages/client/src/stores/hermes/chat.ts

@@ -562,6 +562,7 @@ function mapHermesSession(s: SessionSummary): Session {
     updatedAt: Math.round(activitySeconds * 1000),
     model: s.model,
     provider: s.provider || (s as any).billing_provider || '',
+    apiMode: s.api_mode,
     messageCount: s.message_count,
     messageTotal: s.message_count,
     loadedMessageCount: 0,
@@ -615,6 +616,10 @@ function clearCodingAgentRuntimeCredentials(session?: Session | null) {
   session.apiKey = undefined
 }
 
+function shouldPreserveRuntimeApiMode(session?: Session | null): boolean {
+  return isCodingAgentLikeSession(session) && session?.codingAgentMode !== 'global'
+}
+
 function isQuotaExceededError(error: unknown): boolean {
   if (!error || typeof error !== 'object') return false
   const e = error as { name?: string, code?: number }
@@ -982,6 +987,7 @@ export const useChatStore = defineStore('chat', () => {
         ...mapped,
         messages: existing.messages,
         contextTokens: existing.contextTokens,
+        apiMode: mapped.apiMode || existing.apiMode,
         loadedMessageCount: existing.loadedMessageCount,
         hasMoreBefore: existing.hasMoreBefore,
       })
@@ -1001,11 +1007,13 @@ export const useChatStore = defineStore('chat', () => {
       const runtimeByIdBefore = new Map(sessions.value.map(s => [s.id, {
         messages: s.messages,
         contextTokens: s.contextTokens,
+        apiMode: s.apiMode,
       }]))
       for (const s of fresh) {
         const prev = runtimeByIdBefore.get(s.id)
         if (prev?.messages?.length) s.messages = prev.messages
         if (prev?.contextTokens != null) s.contextTokens = prev.contextTokens
+        if (!s.apiMode && prev?.apiMode) s.apiMode = prev.apiMode
       }
       sessions.value = fresh
       pruneCompletedUnreadSessions(new Set(sessions.value.map(s => s.id)))
@@ -1073,6 +1081,7 @@ export const useChatStore = defineStore('chat', () => {
           existing.endedAt = fresh.endedAt
           existing.model = fresh.model
           existing.provider = fresh.provider
+          existing.apiMode = fresh.apiMode || existing.apiMode
           existing.messageCount = fresh.messageCount
           existing.inputTokens = fresh.inputTokens
           existing.outputTokens = fresh.outputTokens
@@ -1468,18 +1477,22 @@ export const useChatStore = defineStore('chat', () => {
     const shouldClearRuntimeCredentials = previousProvider !== nextProvider && (
       isCodingAgentLikeSession(target) || isCodingAgentLikeSession(activeTarget)
     )
-    const ok = await setSessionModel(targetId, modelId, provider || '', apiMode)
+    const preservedApiMode = apiMode || (previousProvider === nextProvider
+      ? (shouldPreserveRuntimeApiMode(target) ? target?.apiMode : undefined) ||
+        (shouldPreserveRuntimeApiMode(activeTarget) ? activeTarget?.apiMode : undefined)
+      : undefined)
+    const ok = await setSessionModel(targetId, modelId, provider || '', preservedApiMode)
     if (!ok) return false
     if (target) {
       target.model = modelId
       target.provider = provider || ''
-      if (apiMode) target.apiMode = apiMode
+      target.apiMode = preservedApiMode
       if (shouldClearRuntimeCredentials) clearCodingAgentRuntimeCredentials(target)
     }
     if (activeTarget) {
       activeTarget.model = modelId
       activeTarget.provider = provider || ''
-      if (apiMode) activeTarget.apiMode = apiMode
+      activeTarget.apiMode = preservedApiMode
       if (shouldClearRuntimeCredentials) clearCodingAgentRuntimeCredentials(activeTarget)
     }
     return true

packages/server/src/controllers/hermes/sessions.ts

@@ -323,6 +323,7 @@ export async function listConversations(ctx: any) {
     agent_native_session_id: s.agent_native_session_id,
     model: s.model,
     provider: s.provider,
+    api_mode: s.api_mode,
     title: s.title,
     started_at: s.started_at,
     ended_at: s.ended_at,
@@ -985,8 +986,17 @@ export async function setWorkspace(ctx: any) {
   ctx.body = { ok: true }
 }
 
+type SessionProviderApiMode = 'chat_completions' | 'codex_responses' | 'anthropic_messages'
+
+function normalizeSessionApiMode(value: unknown): SessionProviderApiMode | undefined {
+  const mode = typeof value === 'string' ? value.trim() : ''
+  return mode === 'chat_completions' || mode === 'codex_responses' || mode === 'anthropic_messages'
+    ? mode
+    : undefined
+}
+
 export async function setModel(ctx: any) {
-  const { model, provider } = ctx.request.body as { model?: string; provider?: string }
+  const { model, provider, apiMode, api_mode } = ctx.request.body as { model?: string; provider?: string; apiMode?: SessionProviderApiMode; api_mode?: SessionProviderApiMode }
   if (!model || typeof model !== 'string') {
     ctx.status = 400
     ctx.body = { error: 'model is required' }
@@ -1004,19 +1014,22 @@ export async function setModel(ctx: any) {
   const profile = existing?.profile || requestedProfile(ctx) || 'default'
   const cleanModel = model.trim()
   const cleanProvider = (provider || '').trim()
+  const cleanApiMode = normalizeSessionApiMode(apiMode ?? api_mode)
   const codingAgentSession = isCodingAgentSession(existing)
   const workspace = !codingAgentSession
     ? await ensureHermesRunWorkspace(profile, existing?.workspace)
     : undefined
   if (!existing) {
-    createSession({ id, profile, title: '', model: cleanModel, provider: cleanProvider, workspace })
+    createSession({ id, profile, title: '', model: cleanModel, provider: cleanProvider, api_mode: cleanApiMode || '', workspace })
   }
   const updates: Record<string, string> = { model: cleanModel, provider: cleanProvider }
+  if (cleanApiMode) updates.api_mode = cleanApiMode
+  else if (codingAgentSession && existing && existing.provider !== cleanProvider) updates.api_mode = ''
   if (!codingAgentSession && existing && !existing.workspace && workspace) updates.workspace = workspace
   if (
     codingAgentSession &&
     existing &&
-    (existing.model !== cleanModel || existing.provider !== cleanProvider)
+    (existing.model !== cleanModel || existing.provider !== cleanProvider || (cleanApiMode && existing.api_mode !== cleanApiMode))
   ) {
     updates.agent_native_session_id = ''
   }

packages/server/src/db/hermes/schemas.ts

@@ -39,6 +39,7 @@ export const SESSIONS_SCHEMA: Record<string, string> = {
   user_id: 'TEXT',
   model: 'TEXT NOT NULL DEFAULT \'\'',
   provider: 'TEXT NOT NULL DEFAULT \'\'',
+  api_mode: 'TEXT NOT NULL DEFAULT \'\'',
   title: 'TEXT',
   parent_session_id: 'TEXT',
   fork_point_message_id: 'TEXT',

packages/server/src/db/hermes/session-store.ts

@@ -19,6 +19,7 @@ export interface HermesSessionRow {
   user_id: string | null
   model: string
   provider: string
+  api_mode: string
   title: string | null
   parent_session_id: string | null
   fork_point_message_id: string | null
@@ -105,6 +106,7 @@ function mapSessionRow(row: Record<string, unknown>): HermesSessionRow {
     user_id: row.user_id != null ? String(row.user_id) : null,
     model: String(row.model || ''),
     provider: String(row.provider || ''),
+    api_mode: String(row.api_mode || ''),
     title,
     parent_session_id: row.parent_session_id != null ? String(row.parent_session_id) : null,
     fork_point_message_id: row.fork_point_message_id != null ? String(row.fork_point_message_id) : null,
@@ -165,6 +167,7 @@ export function createSession(data: {
   agent_native_session_id?: string
   model?: string
   provider?: string
+  api_mode?: string
   title?: string
   parent_session_id?: string | null
   workspace?: string
@@ -177,7 +180,7 @@ export function createSession(data: {
       id: data.id, profile: data.profile || 'default', source, agent,
       agent_mode: data.agent_mode || '',
       agent_session_id: data.agent_session_id || '', agent_native_session_id: data.agent_native_session_id || '',
-      user_id: null, model: data.model || '', provider: data.provider || '', title: data.title || null,
+      user_id: null, model: data.model || '', provider: data.provider || '', api_mode: data.api_mode || '', title: data.title || null,
       parent_session_id: data.parent_session_id || null,
       fork_point_message_id: null,
       started_at: now, ended_at: null, end_reason: null,
@@ -189,8 +192,8 @@ export function createSession(data: {
   }
   const db = getDb()!
   db.prepare(
-    `INSERT INTO ${SESSIONS_TABLE} (id, profile, source, agent, agent_mode, agent_session_id, agent_native_session_id, model, provider, title, parent_session_id, started_at, last_active, workspace)
-     VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`,
+    `INSERT INTO ${SESSIONS_TABLE} (id, profile, source, agent, agent_mode, agent_session_id, agent_native_session_id, model, provider, api_mode, title, parent_session_id, started_at, last_active, workspace)
+     VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`,
   ).run(
     data.id,
     data.profile || 'default',
@@ -201,6 +204,7 @@ export function createSession(data: {
     data.agent_native_session_id || '',
     data.model || '',
     data.provider || '',
+    data.api_mode || '',
     data.title || null,
     data.parent_session_id || null,
     now,
@@ -221,6 +225,7 @@ export function createBranchedSession(data: {
   agent_native_session_id?: string
   model?: string
   provider?: string
+  api_mode?: string
   title?: string
   workspace?: string | null
   ended_at: number
@@ -257,8 +262,8 @@ export function createBranchedSession(data: {
     ).run(data.ended_at, 'branched', data.parent_session_id)
 
     db.prepare(
-      `INSERT INTO ${SESSIONS_TABLE} (id, profile, source, agent, agent_mode, agent_session_id, agent_native_session_id, model, provider, title, parent_session_id, started_at, last_active, workspace, message_count)
-       VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`,
+      `INSERT INTO ${SESSIONS_TABLE} (id, profile, source, agent, agent_mode, agent_session_id, agent_native_session_id, model, provider, api_mode, title, parent_session_id, started_at, last_active, workspace, message_count)
+       VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`,
     ).run(
       data.id,
       data.profile || 'default',
@@ -269,6 +274,7 @@ export function createBranchedSession(data: {
       data.agent_native_session_id || '',
       data.model || '',
       data.provider || '',
+      data.api_mode || '',
       data.title || null,
       data.parent_session_id,
       data.ended_at,

packages/server/src/services/agent-runner/coding-agent-run-manager.ts

@@ -3,6 +3,7 @@ import { existsSync, accessSync, chmodSync, constants as fsConstants, readFileSy
 import { homedir } from 'os'
 import { spawn, type ChildProcess } from 'child_process'
 import { createSession, addMessage, getSession, updateSession, updateSessionStats } from '../../db/hermes/session-store'
+import type { ApiMode } from './types'
 import { logger } from '../logger'
 import { applyResponseStreamEvent, flushResponseRunToDb } from '../hermes/run-chat/response-stream'
 import { calcAndUpdateUsage, updateMessageContextTokenUsage } from '../hermes/run-chat/usage'
@@ -62,6 +63,7 @@ export interface CodingAgentRunLaunch {
   profile: string
   provider: string
   model: string
+  apiMode?: ApiMode
   sessionId: string
   agentNativeSessionId?: string
   nativeResume?: boolean
@@ -417,6 +419,7 @@ export class CodingAgentRunManager {
     provider?: string
     model?: string
     reasoningEffort?: string
+    apiMode?: ApiMode
   }): boolean {
     const run = this.getBySession(sessionId)
     if (!run || run.exited) return false
@@ -428,6 +431,8 @@ export class CodingAgentRunManager {
       const model = String(launch.model || '').trim()
       if (provider && run.launch.provider !== provider) return false
       if (model && run.launch.model !== model) return false
+      const apiMode = String(launch.apiMode || '').trim()
+      if (apiMode && String(run.launch.apiMode || '').trim() !== apiMode) return false
     }
     if (String(run.launch.reasoningEffort || '').trim() !== String(launch.reasoningEffort || '').trim()) return false
     if (!hasManagedHermesMcpConfig(run)) return false
@@ -697,12 +702,13 @@ export class CodingAgentRunManager {
     createSession({
       id: run.launch.sessionId,
       profile: run.launch.profile,
-        source,
-        agent: run.launch.agentId === 'codex' ? 'codex' : 'claude',
-        agent_session_id: run.id,
-        agent_native_session_id: run.launch.agentNativeSessionId,
-        model: run.launch.model,
+      source,
+      agent: run.launch.agentId === 'codex' ? 'codex' : 'claude',
+      agent_session_id: run.id,
+      agent_native_session_id: run.launch.agentNativeSessionId,
+      model: run.launch.model,
       provider: run.launch.provider,
+      api_mode: run.launch.apiMode || '',
       title: '',
       workspace: run.launch.workspaceDir,
     })

packages/server/src/services/coding-agents.ts

@@ -124,6 +124,7 @@ export interface CodingAgentLaunchResult {
   profile: string
   provider: string
   model: string
+  apiMode?: ApiMode
   rootDir: string
   workspaceDir: string
   command: string
@@ -500,12 +501,17 @@ async function resolveStoredProviderLaunchInput(
   if (input.mode === 'global') return input
 
   const profile = String(input.profile || existingSession?.profile || 'default').trim() || 'default'
-  const provider = String(input.provider || existingSession?.provider || '').trim()
+  const inputProvider = String(input.provider || '').trim()
+  const storedProvider = String(existingSession?.provider || '').trim()
+  const provider = String(inputProvider || storedProvider).trim()
   const model = String(input.model || existingSession?.model || '').trim()
   const workspace = input.workspace || existingSession?.workspace || undefined
   let baseUrl = String(input.baseUrl || '').trim()
   let apiKey = String(input.apiKey || '').trim()
-  let apiMode = input.apiMode
+  const storedApiMode = !inputProvider || inputProvider === storedProvider
+    ? normalizeStoredLaunchApiMode(existingSession?.api_mode)
+    : undefined
+  let apiMode = input.apiMode || storedApiMode
   let canonicalProvider = provider
   const ignoredStaleProviderRuntime = belongsToDifferentBuiltinProvider(provider, baseUrl)
   if (ignoredStaleProviderRuntime) {
@@ -577,6 +583,15 @@ async function resolveStoredProviderLaunchInput(
   }
 }
 
+function normalizeStoredLaunchApiMode(value: unknown): ApiMode | undefined {
+  if (!value) return undefined
+  try {
+    return normalizeLaunchApiMode(value, 'chat_completions')
+  } catch {
+    return undefined
+  }
+}
+
 function normalizeLaunchApiMode(value: unknown, fallback: ApiMode): ApiMode {
   if (!value) return fallback
   const mode = String(value).trim() as ApiMode
@@ -1726,6 +1741,7 @@ export async function prepareCodingAgentLaunch(id: string, input: CodingAgentLau
     profile: scope.profile,
     provider: scope.provider,
     model,
+    apiMode,
     rootDir,
     workspaceDir,
     command: tool.command,
@@ -1773,7 +1789,8 @@ export async function startCodingAgentRun(
     ? storedCodingAgentMode(existingSession) === requestedMode &&
       (existingSession.agent === (id === 'codex' ? 'codex' : 'claude') || !existingSession.agent) &&
       String(existingSession.provider || '').trim() === String(resolvedInput.provider || '').trim() &&
-      String(existingSession.model || '').trim() === String(resolvedInput.model || '').trim()
+      String(existingSession.model || '').trim() === String(resolvedInput.model || '').trim() &&
+      (!String(existingSession.api_mode || '').trim() || String(existingSession.api_mode || '').trim() === String(resolvedInput.apiMode || '').trim())
     : false
   const existingNativeSessionId = canResumeNativeSession ? existingSession?.agent_native_session_id || '' : ''
   const agentNativeSessionId = resolvedInput.agentNativeSessionId || existingNativeSessionId || (id === 'claude-code' ? randomUUID() : '')
@@ -1800,6 +1817,7 @@ export async function startCodingAgentRun(
     profile: launch.profile,
     provider: persistedProvider,
     model: launch.model,
+    apiMode: launch.apiMode,
     sessionId,
     agentNativeSessionId,
     nativeResume: Boolean(existingNativeSessionId),
@@ -1820,6 +1838,7 @@ export async function startCodingAgentRun(
     agent_native_session_id: agentNativeSessionId,
     model: launch.model,
     provider: persistedProvider,
+    api_mode: launch.apiMode || '',
     workspace: launch.workspaceDir,
   })
   return {

packages/server/src/services/hermes/conversations.ts

@@ -42,6 +42,7 @@ export interface ConversationSummary {
   agent_native_session_id?: string
   model: string
   provider?: string
+  api_mode?: string
   title: string | null
   started_at: number
   ended_at: number | null

packages/server/src/services/hermes/run-chat/handle-coding-agent-run.ts

@@ -64,11 +64,13 @@ export async function handleCodingAgentRun(
   const storedSession = getSession(sessionId)
   const launchProvider = data.provider || (mode === 'scoped' ? storedSession?.provider || undefined : undefined)
   const launchModel = data.model || (mode === 'scoped' ? storedSession?.model || undefined : undefined)
+  const launchApiMode = data.apiMode || data.api_mode || (mode === 'scoped' ? storedSession?.api_mode || undefined : undefined)
   if (runId && !codingAgentRunManager.isSessionLaunchCompatible(sessionId, {
     agentId,
     mode,
     provider: launchProvider,
     model: launchModel,
+    apiMode: launchApiMode,
     reasoningEffort: data.reasoning_effort,
   })) {
     codingAgentRunManager.stop(sessionId, { reportClosed: false })
@@ -84,7 +86,7 @@ export async function handleCodingAgentRun(
       workspace: data.workspace,
       baseUrl: data.baseUrl || data.base_url,
       apiKey: data.apiKey || data.api_key,
-      apiMode: data.apiMode || data.api_mode,
+      apiMode: launchApiMode,
       reasoningEffort: data.reasoning_effort,
       sessionSource: data.session_source,
     }, state)

packages/server/src/services/hermes/run-chat/session-command.ts

@@ -1117,6 +1117,7 @@ function createBranchSession(parentSessionId: string, requestedTitle: string, ct
     agent_native_session_id: parent.agent_native_session_id || '',
     model: parent.model || ctx.model || '',
     provider: parent.provider || ctx.provider || '',
+    api_mode: parent.api_mode || '',
     title,
     parent_session_id: parentSessionId,
     workspace: parent.workspace || undefined,

scripts/generate-openapi.mjs

@@ -608,6 +608,9 @@ function schemaFromType(type) {
   const schema = {}
 
   if (/\bnull\b/.test(normalized)) schema.nullable = true
+  if (/SessionProviderApiMode|CodingAgentApiMode/.test(normalized)) {
+    return { ...schema, type: 'string', enum: ['chat_completions', 'codex_responses', 'anthropic_messages'] }
+  }
   if (/string\[\]|Array<string>/.test(normalized)) {
     return { ...schema, type: 'array', items: { type: 'string' } }
   }

tests/client/chat-store-reasoning-tool-boundary.test.ts

@@ -546,6 +546,33 @@ describe('chat store reasoning/tool boundaries', () => {
     expect(session.apiMode).toBe('chat_completions')
   })
 
+
+  it('preserves a scoped coding-agent API mode when reselecting the same provider', async () => {
+    const store = useChatStore()
+    const session = makeSession()
+    session.source = 'coding_agent'
+    session.agent = 'codex'
+    session.codingAgentId = 'codex'
+    session.codingAgentMode = 'scoped'
+    session.provider = 'fun-codex'
+    session.model = 'gpt-5.4'
+    session.apiMode = 'chat_completions'
+    store.sessions = [session]
+    store.activeSessionId = 'session-1'
+    store.activeSession = session
+
+    const ok = await store.switchSessionModel('gpt-5.5', 'fun-codex', 'session-1')
+
+    expect(ok).toBe(true)
+    expect(sessionsApi.setSessionModel).toHaveBeenCalledWith(
+      'session-1',
+      'gpt-5.5',
+      'fun-codex',
+      'chat_completions',
+    )
+    expect(session.apiMode).toBe('chat_completions')
+  })
+
   it('sends the selected workspace when starting a coding-agent run', async () => {
     const store = useChatStore()
     const session = makeSession()

tests/server/coding-agent-resume-config.test.ts

@@ -278,6 +278,39 @@ describe('coding agent resumed session config', () => {
     }))
   })
 
+
+  it('prefers a stored coding-agent API mode when resuming without an explicit mode', async () => {
+    const home = makeHome()
+    getSessionMock.mockReturnValue({
+      id: 'session-1',
+      profile: 'default',
+      source: 'coding_agent',
+      agent: 'codex',
+      agent_session_id: 'agent-session-1',
+      provider: 'fun-codex',
+      model: 'gpt-5.5',
+      api_mode: 'chat_completions',
+    })
+    readConfigYamlForProfileMock.mockResolvedValue({})
+    safeReadFileMock.mockResolvedValue('')
+
+    const { startCodingAgentRun } = await import('../../packages/server/src/services/coding-agents')
+    await startCodingAgentRun('codex', {
+      sessionId: 'session-1',
+      baseUrl: 'https://api.apikey.fun/v1',
+      apiKey: 'sk-test',
+    })
+
+    const launch = startRunMock.mock.calls[0][0]
+    expect(launch.apiMode).toBe('chat_completions')
+    const config = readFileSync(join(home, 'coding-agent', 'model', 'default', 'fun-codex', 'codex', 'config.toml'), 'utf-8')
+    const routeKey = config.match(/\/api\/codex-proxy\/([^/]+)\/v1/)?.[1] || ''
+    expect(Buffer.from(routeKey, 'base64url').toString('utf8')).toContain('chat_completions')
+    expect(updateSessionMock).toHaveBeenCalledWith('session-1', expect.objectContaining({
+      api_mode: 'chat_completions',
+    }))
+  })
+
   it('rejects OAuth/subscription providers for scoped coding-agent launches', async () => {
     makeHome()
     getSessionMock.mockReturnValue(null)

tests/server/session-command-branch.test.ts

@@ -83,6 +83,7 @@ function makeParentSession(overrides: Record<string, any> = {}) {
     agent_native_session_id: '',
     model: 'openai/gpt-5.4',
     provider: 'openai-codex',
+    api_mode: 'chat_completions',
     title: 'Parent chat',
     preview: 'Parent prompt',
     workspace: '/repo',
@@ -229,6 +230,7 @@ describe('branch session command', () => {
       agent: 'hermes',
       model: 'openai/gpt-5.4',
       provider: 'openai-codex',
+      api_mode: 'chat_completions',
       workspace: '/repo',
       ended_at: expect.any(Number),
       last_active: expect.any(Number),

tests/server/session-store-branch.test.ts

@@ -52,6 +52,7 @@ describe('createBranchedSession', () => {
       user_id: null,
       model: 'openai/gpt-5.4',
       provider: 'openai-codex',
+      api_mode: 'chat_completions',
       title: 'Forked child',
       parent_session_id: 'parent-session',
       fork_point_message_id: '42',
@@ -86,6 +87,7 @@ describe('createBranchedSession', () => {
       agent: 'hermes',
       model: 'openai/gpt-5.4',
       provider: 'openai-codex',
+      api_mode: 'chat_completions',
       title: 'Forked child',
       workspace: '/repo',
       ended_at: 123,

tests/server/sessions-controller.test.ts

@@ -1317,7 +1317,7 @@ describe('session conversations controller', () => {
     expect(ctx.body).toEqual({ ok: true })
   })
 
-  it('stores a coding agent session model without stopping the runner or notifying the Hermes bridge', async () => {
+  it('stores a coding agent session model and API mode without stopping the runner or notifying the Hermes bridge', async () => {
     bridgeGetRuntimeStateMock.mockReturnValue({ ready: true, running: true, endpoint: 'ipc:///tmp/hermes-agent-bridge.sock' })
     getSessionMock.mockReturnValue({
       id: 'codex-session',
@@ -1326,6 +1326,7 @@ describe('session conversations controller', () => {
       agent: 'codex',
       model: 'old-model',
       provider: 'openrouter',
+      api_mode: 'codex_responses',
       agent_native_session_id: 'old-native-thread',
       workspace: '/tmp/original-workspace',
     })
@@ -1333,7 +1334,7 @@ describe('session conversations controller', () => {
     const mod = await import('../../packages/server/src/controllers/hermes/sessions')
     const ctx: any = {
       params: { id: 'codex-session' },
-      request: { body: { model: 'gpt-5.5', provider: 'openai-codex' } },
+      request: { body: { model: 'gpt-5.5', provider: 'openai-codex', apiMode: 'chat_completions' } },
       body: null,
     }
     await mod.setModel(ctx)
@@ -1341,6 +1342,7 @@ describe('session conversations controller', () => {
     expect(localUpdateSessionMock).toHaveBeenCalledWith('codex-session', {
       model: 'gpt-5.5',
       provider: 'openai-codex',
+      api_mode: 'chat_completions',
       agent_native_session_id: '',
     })
     expect(codingAgentRunManagerMock.stop).not.toHaveBeenCalled()