Add DeepSeek V4 provider

Author: jahooma

agents/base2/base2-free-deepseek-v4.ts

@@ -0,0 +1,11 @@
+import { createBase2 } from './base2'
+
+const definition = {
+  ...createBase2('free', {
+    noAskUser: true,
+    model: 'deepseek/deepseek-v4-pro',
+  }),
+  id: 'base2-free-deepseek-v4',
+  displayName: 'Buffy the DeepSeek V4 Free Orchestrator',
+}
+export default definition

agents/types/agent-definition.ts

@@ -415,6 +415,8 @@ export type ModelName =
   | 'qwen/qwen3-30b-a3b:nitro'
 
   // DeepSeek
+  | 'deepseek/deepseek-v4-pro'
+  | 'deepseek-v4-pro'
   | 'deepseek/deepseek-chat-v3-0324'
   | 'deepseek/deepseek-chat-v3-0324:nitro'
   | 'deepseek/deepseek-r1-0528'

common/src/constants/free-agents.ts

@@ -1,6 +1,9 @@
 import { parseAgentId } from '../util/agent-id-parsing'
 
-import { SUPPORTED_FREEBUFF_MODELS } from './freebuff-models'
+import {
+  FREEBUFF_DEEPSEEK_V4_PRO_MODEL_ID,
+  SUPPORTED_FREEBUFF_MODELS,
+} from './freebuff-models'
 
 import type { CostMode } from './model-config'
 
@@ -16,7 +19,10 @@ export const FREE_COST_MODE = 'free' as const
  * excluded — they're spawned by the root, so counting them would inflate
  * every user's apparent activity.
  */
-export const FREEBUFF_ROOT_AGENT_IDS = ['base2-free'] as const
+export const FREEBUFF_ROOT_AGENT_IDS = [
+  'base2-free',
+  'base2-free-deepseek-v4',
+] as const
 const FREEBUFF_ROOT_AGENT_ID_SET: ReadonlySet<string> = new Set(
   FREEBUFF_ROOT_AGENT_IDS,
 )
@@ -35,6 +41,7 @@ const FREEBUFF_ALLOWED_MODEL_IDS = SUPPORTED_FREEBUFF_MODELS.map(
 export const FREE_MODE_AGENT_MODELS: Record<string, Set<string>> = {
   // Root orchestrator
   'base2-free': new Set(FREEBUFF_ALLOWED_MODEL_IDS),
+  'base2-free-deepseek-v4': new Set([FREEBUFF_DEEPSEEK_V4_PRO_MODEL_ID]),
 
   // File exploration agents
   'file-picker': new Set(['google/gemini-2.5-flash-lite']),

common/src/constants/freebuff-models.ts

@@ -22,6 +22,7 @@ export interface FreebuffModelOption {
  *  `getFreebuffDeploymentAvailabilityLabel()` instead. */
 export const FREEBUFF_DEPLOYMENT_HOURS_LABEL = '9am ET-5pm PT every day'
 export const FREEBUFF_GEMINI_PRO_MODEL_ID = 'google/gemini-3.1-pro-preview'
+export const FREEBUFF_DEEPSEEK_V4_PRO_MODEL_ID = 'deepseek/deepseek-v4-pro'
 export const FREEBUFF_GLM_MODEL_ID = 'z-ai/glm-5.1'
 export const FREEBUFF_KIMI_MODEL_ID = 'moonshotai/kimi-k2.6'
 export const FREEBUFF_MINIMAX_MODEL_ID = 'minimax/minimax-m2.7'
@@ -48,6 +49,12 @@ export const FREEBUFF_MODELS = [
     tagline: 'Deepest, 1/day',
     availability: 'always',
   },
+  {
+    id: FREEBUFF_DEEPSEEK_V4_PRO_MODEL_ID,
+    displayName: 'DeepSeek V4 Pro',
+    tagline: 'Experimental',
+    availability: 'always',
+  },
   {
     id: FREEBUFF_MINIMAX_MODEL_ID,
     displayName: 'MiniMax M2.7',

common/src/constants/model-config.ts

@@ -6,6 +6,7 @@ export const ALLOWED_MODEL_PREFIXES = [
   'openai',
   'google',
   'x-ai',
+  'deepseek',
 ] as const
 
 export const costModes = [
@@ -55,6 +56,8 @@ export type openrouterModel =
 export const deepseekModels = {
   deepseekChat: 'deepseek-chat',
   deepseekReasoner: 'deepseek-reasoner',
+  deepseekV4ProDirect: 'deepseek-v4-pro',
+  deepseekV4Pro: 'deepseek/deepseek-v4-pro',
 } as const
 export type DeepseekModel = (typeof deepseekModels)[keyof typeof deepseekModels]
 

common/src/templates/initial-agents-dir/types/agent-definition.ts

@@ -415,6 +415,8 @@ export type ModelName =
   | 'qwen/qwen3-30b-a3b:nitro'
 
   // DeepSeek
+  | 'deepseek/deepseek-v4-pro'
+  | 'deepseek-v4-pro'
   | 'deepseek/deepseek-chat-v3-0324'
   | 'deepseek/deepseek-chat-v3-0324:nitro'
   | 'deepseek/deepseek-r1-0528'

evals/buffbench/main-single-eval.ts

@@ -7,7 +7,7 @@ async function main() {
 
   await runBuffBench({
     evalDataPaths: [path.join(__dirname, 'eval-codebuff.json')],
-    agents: ['base2-free-evals'],
+    agents: ['base2-free-deepseek-v4'],
     taskIds: ['server-agent-validation'],
     saveTraces,
   })

packages/internal/src/env-schema.ts

@@ -8,6 +8,7 @@ export const serverEnvSchema = clientEnvSchema.extend({
   ANTHROPIC_API_KEY: z.string().min(1),
   FIREWORKS_API_KEY: z.string().min(1),
   CANOPYWAVE_API_KEY: z.string().min(1).optional(),
+  DEEPSEEK_API_KEY: z.string().min(1).optional(),
   SILICONFLOW_API_KEY: z.string().min(1).optional(),
   LINKUP_API_KEY: z.string().min(1),
   CONTEXT7_API_KEY: z.string().optional(),
@@ -92,6 +93,7 @@ export const serverProcessEnv: ServerInput = {
   ANTHROPIC_API_KEY: process.env.ANTHROPIC_API_KEY,
   FIREWORKS_API_KEY: process.env.FIREWORKS_API_KEY,
   CANOPYWAVE_API_KEY: process.env.CANOPYWAVE_API_KEY,
+  DEEPSEEK_API_KEY: process.env.DEEPSEEK_API_KEY,
   SILICONFLOW_API_KEY: process.env.SILICONFLOW_API_KEY,
   LINKUP_API_KEY: process.env.LINKUP_API_KEY,
   CONTEXT7_API_KEY: process.env.CONTEXT7_API_KEY,

packages/internal/src/env.ts

@@ -18,6 +18,7 @@ if (isCI) {
   ensureEnvDefault('ANTHROPIC_API_KEY', 'test')
   ensureEnvDefault('FIREWORKS_API_KEY', 'test')
   ensureEnvDefault('CANOPYWAVE_API_KEY', 'test')
+  ensureEnvDefault('DEEPSEEK_API_KEY', 'test')
   ensureEnvDefault('LINKUP_API_KEY', 'test')
   ensureEnvDefault('GRAVITY_API_KEY', 'test')
   ensureEnvDefault('IPINFO_TOKEN', 'test')

web/src/app/api/v1/chat/completions/__tests__/completions.test.ts

@@ -2,6 +2,7 @@ import { afterEach, beforeEach, describe, expect, mock, it } from 'bun:test'
 import { NextRequest } from 'next/server'
 
 import {
+  FREEBUFF_DEEPSEEK_V4_PRO_MODEL_ID,
   FREEBUFF_GEMINI_PRO_MODEL_ID,
   FREEBUFF_GLM_MODEL_ID,
   isFreebuffDeploymentHours,
@@ -147,6 +148,13 @@ describe('/api/v1/chat/completions POST endpoint', () => {
           status: 'running',
         }
       }
+      if (runId === 'run-free-deepseek-v4') {
+        return {
+          agent_id: 'base2-free-deepseek-v4',
+          ancestor_run_ids: [],
+          status: 'running',
+        }
+      }
       if (runId === 'run-reviewer-direct') {
         return {
           agent_id: 'code-reviewer-lite',
@@ -823,6 +831,111 @@ describe('/api/v1/chat/completions POST endpoint', () => {
       FETCH_PATH_TEST_TIMEOUT_MS,
     )
 
+    it(
+      'lets the DeepSeek V4 free agent use the direct DeepSeek provider',
+      async () => {
+        const fetchedBodies: Record<string, unknown>[] = []
+        const fetchedUrls: string[] = []
+        const fetchViaDeepSeek = mock(
+          async (url: string | URL | Request, init?: RequestInit) => {
+            fetchedUrls.push(String(url))
+            fetchedBodies.push(JSON.parse(init?.body as string))
+            return new Response(
+              JSON.stringify({
+                id: 'test-id',
+                model: 'deepseek-v4-pro',
+                choices: [{ message: { content: 'test response' } }],
+                usage: {
+                  prompt_tokens: 10,
+                  prompt_cache_hit_tokens: 4,
+                  completion_tokens: 20,
+                  total_tokens: 30,
+                },
+              }),
+              {
+                status: 200,
+                headers: { 'Content-Type': 'application/json' },
+              },
+            )
+          },
+        ) as unknown as typeof globalThis.fetch
+
+        const req = new NextRequest(
+          'http://localhost:3000/api/v1/chat/completions',
+          {
+            method: 'POST',
+            headers: allowedFreeModeHeaders('test-api-key-new-free'),
+            body: JSON.stringify({
+              model: FREEBUFF_DEEPSEEK_V4_PRO_MODEL_ID,
+              stream: false,
+              codebuff_metadata: {
+                run_id: 'run-free-deepseek-v4',
+                client_id: 'test-client-id-123',
+                cost_mode: 'free',
+              },
+            }),
+          },
+        )
+
+        const response = await postChatCompletions({
+          req,
+          getUserInfoFromApiKey: mockGetUserInfoFromApiKey,
+          logger: mockLogger,
+          trackEvent: mockTrackEvent,
+          getUserUsageData: mockGetUserUsageData,
+          getAgentRunFromId: mockGetAgentRunFromId,
+          fetch: fetchViaDeepSeek,
+          insertMessageBigquery: mockInsertMessageBigquery,
+          loggerWithContext: mockLoggerWithContext,
+          checkSessionAdmissible: mockCheckSessionAdmissibleAllow,
+        })
+
+        const body = await response.json()
+        expect(response.status).toBe(200)
+        expect(fetchedUrls[0]).toBe('https://api.deepseek.com/chat/completions')
+        expect(fetchedBodies[0].model).toBe('deepseek-v4-pro')
+        expect(body.model).toBe(FREEBUFF_DEEPSEEK_V4_PRO_MODEL_ID)
+        expect(body.provider).toBe('DeepSeek')
+      },
+      FETCH_PATH_TEST_TIMEOUT_MS,
+    )
+
+    it('rejects the DeepSeek V4 free agent when it requests another free model', async () => {
+      const req = new NextRequest(
+        'http://localhost:3000/api/v1/chat/completions',
+        {
+          method: 'POST',
+          headers: allowedFreeModeHeaders('test-api-key-new-free'),
+          body: JSON.stringify({
+            model: FREEBUFF_GEMINI_PRO_MODEL_ID,
+            stream: false,
+            codebuff_metadata: {
+              run_id: 'run-free-deepseek-v4',
+              client_id: 'test-client-id-123',
+              cost_mode: 'free',
+            },
+          }),
+        },
+      )
+
+      const response = await postChatCompletions({
+        req,
+        getUserInfoFromApiKey: mockGetUserInfoFromApiKey,
+        logger: mockLogger,
+        trackEvent: mockTrackEvent,
+        getUserUsageData: mockGetUserUsageData,
+        getAgentRunFromId: mockGetAgentRunFromId,
+        fetch: mockFetch,
+        insertMessageBigquery: mockInsertMessageBigquery,
+        loggerWithContext: mockLoggerWithContext,
+        checkSessionAdmissible: mockCheckSessionAdmissibleAllow,
+      })
+
+      const body = await response.json()
+      expect(response.status).toBe(403)
+      expect(body.error).toBe('free_mode_invalid_agent_model')
+    })
+
     it('lets freebuff use Gemini 3.1 Pro through the free-mode allowlist', async () => {
       const req = new NextRequest(
         'http://localhost:3000/api/v1/chat/completions',