diff --git a/common/src/constants/requesty.ts b/common/src/constants/requesty.ts
new file mode 100644
index 0000000000..a8421599e4
--- /dev/null
+++ b/common/src/constants/requesty.ts
@@ -0,0 +1,16 @@
+/**
+ * Constants for the optional direct Requesty provider route.
+ *
+ * Requesty (https://requesty.ai) exposes an OpenAI-compatible router at
+ * https://router.requesty.ai/v1. When `REQUESTY_API_KEY` is set, the SDK can
+ * route chat completions directly to Requesty using the same OpenAI-compatible
+ * language model shim used elsewhere, instead of going through the Codebuff
+ * backend. Model ids use the same `provider/model` form as OpenRouter
+ * (e.g. `openai/gpt-4o-mini`, `anthropic/claude-sonnet-4.5`).
+ */
+
+/** Base URL for the Requesty OpenAI-compatible router. */
+export const REQUESTY_BASE_URL = 'https://router.requesty.ai/v1'
+
+/** Environment variable holding the Requesty API key. */
+export const REQUESTY_ENV_VAR = 'REQUESTY_API_KEY'
diff --git a/docs/agents-and-tools.md b/docs/agents-and-tools.md
index 4ea7475896..f95147def3 100644
--- a/docs/agents-and-tools.md
+++ b/docs/agents-and-tools.md
@@ -19,3 +19,25 @@ base-lite "fix this bug"
 
 - Tool definitions live in `common/src/tools` and are executed via the SDK helpers + agent-runtime.
 
+## Model providers
+
+Model ids use the `provider/model` form (e.g. `openai/gpt-4o-mini`,
+`anthropic/claude-sonnet-4.5`). By default, requests are sent through the
+Codebuff backend, which routes to the upstream provider (OpenRouter).
+
+### Requesty (direct route)
+
+[Requesty](https://requesty.ai) is an OpenAI-compatible router. Setting the
+`REQUESTY_API_KEY` environment variable routes chat completions directly to the
+Requesty router (`https://router.requesty.ai/v1`) instead of the Codebuff
+backend, using the same `provider/model` ids. This mirrors the existing direct
+ChatGPT OAuth route in `sdk/src/impl/model-provider.ts`.
+
+- Router base URL: `https://router.requesty.ai/v1`
+- API keys: https://app.requesty.ai/api-keys
+- Model list: https://app.requesty.ai/router/list
+- Docs: https://docs.requesty.ai
+
+When `REQUESTY_API_KEY` is unset, behavior is unchanged.
+
+
diff --git a/sdk/README.md b/sdk/README.md
index ff7d0ba960..66d99a7ae0 100644
--- a/sdk/README.md
+++ b/sdk/README.md
@@ -262,6 +262,31 @@ The `RunState` object contains:
 - `sessionState`: Internal state to be passed to the next run
 - `output`: The agent's output (text, error, or other types)
 
+## Model providers
+
+By default, the SDK sends requests through the Codebuff backend, which routes to
+the upstream model provider (OpenRouter). Model ids use the `provider/model`
+form, e.g. `openai/gpt-4o-mini` or `anthropic/claude-sonnet-4.5`.
+
+### Requesty (direct, OpenAI-compatible)
+
+[Requesty](https://requesty.ai) exposes an OpenAI-compatible router. When the
+`REQUESTY_API_KEY` environment variable is set, the SDK routes chat completions
+directly to the Requesty router (`https://router.requesty.ai/v1`) instead of the
+Codebuff backend, using the same `provider/model` ids:
+
+```bash
+export REQUESTY_API_KEY="sk-..."   # from https://app.requesty.ai/api-keys
+```
+
+- Router base URL: `https://router.requesty.ai/v1`
+- API keys: https://app.requesty.ai/api-keys
+- Available models: https://app.requesty.ai/router/list
+- Docs: https://docs.requesty.ai
+
+When `REQUESTY_API_KEY` is unset, behavior is unchanged (requests go through the
+Codebuff backend).
+
 ## License
 
 MIT
diff --git a/sdk/src/env.ts b/sdk/src/env.ts
index 033e3f245d..29a81ccd61 100644
--- a/sdk/src/env.ts
+++ b/sdk/src/env.ts
@@ -7,6 +7,7 @@
 
 import { BYOK_OPENROUTER_ENV_VAR } from '@codebuff/common/constants/byok'
 import { CHATGPT_OAUTH_TOKEN_ENV_VAR } from '@codebuff/common/constants/chatgpt-oauth'
+import { REQUESTY_ENV_VAR } from '@codebuff/common/constants/requesty'
 import { API_KEY_ENV_VAR } from '@codebuff/common/constants/paths'
 import { getBaseEnv } from '@codebuff/common/env-process'
 
@@ -42,6 +43,16 @@ export const getByokOpenrouterApiKeyFromEnv = (): string | undefined => {
   return process.env[BYOK_OPENROUTER_ENV_VAR]
 }
 
+/**
+ * Get the Requesty API key from the environment.
+ *
+ * When set, the SDK routes chat completions directly to the Requesty
+ * OpenAI-compatible router instead of going through the Codebuff backend.
+ */
+export const getRequestyApiKeyFromEnv = (): string | undefined => {
+  return process.env[REQUESTY_ENV_VAR]
+}
+
 /**
  * Get ChatGPT OAuth token from environment variable.
  */
diff --git a/sdk/src/impl/__tests__/model-provider-requesty.test.ts b/sdk/src/impl/__tests__/model-provider-requesty.test.ts
new file mode 100644
index 0000000000..07e379df84
--- /dev/null
+++ b/sdk/src/impl/__tests__/model-provider-requesty.test.ts
@@ -0,0 +1,56 @@
+import { describe, expect, test, beforeEach, afterEach } from 'bun:test'
+
+import { REQUESTY_ENV_VAR } from '@codebuff/common/constants/requesty'
+
+describe('getModelForRequest Requesty direct route', () => {
+  const previousRequestyKey = process.env[REQUESTY_ENV_VAR]
+
+  beforeEach(() => {
+    delete process.env[REQUESTY_ENV_VAR]
+  })
+
+  afterEach(() => {
+    if (previousRequestyKey === undefined) {
+      delete process.env[REQUESTY_ENV_VAR]
+    } else {
+      process.env[REQUESTY_ENV_VAR] = previousRequestyKey
+    }
+  })
+
+  async function importFresh() {
+    const mod = await import('../model-provider')
+    mod.resetChatGptOAuthRateLimit()
+    return mod
+  }
+
+  test('routes to Requesty when REQUESTY_API_KEY is set', async () => {
+    process.env[REQUESTY_ENV_VAR] = 'test-requesty-key'
+
+    const { getModelForRequest } = await importFresh()
+
+    const result = await getModelForRequest({
+      apiKey: 'test-codebuff-key',
+      model: 'openai/gpt-4o-mini',
+    })
+
+    expect(result.isChatGptOAuth).toBe(false)
+    expect(typeof result.model).not.toBe('string')
+    if (typeof result.model !== 'string') {
+      expect(result.model.provider).toBe('requesty')
+    }
+  })
+
+  test('uses Codebuff backend when REQUESTY_API_KEY is not set', async () => {
+    const { getModelForRequest } = await importFresh()
+
+    const result = await getModelForRequest({
+      apiKey: 'test-codebuff-key',
+      model: 'openai/gpt-4o-mini',
+    })
+
+    expect(result.isChatGptOAuth).toBe(false)
+    if (typeof result.model !== 'string') {
+      expect(result.model.provider).toBe('codebuff')
+    }
+  })
+})
diff --git a/sdk/src/impl/model-provider.ts b/sdk/src/impl/model-provider.ts
index 268c7394d0..187ff48273 100644
--- a/sdk/src/impl/model-provider.ts
+++ b/sdk/src/impl/model-provider.ts
@@ -10,6 +10,7 @@ import path from 'path'
 
 import { BYOK_OPENROUTER_HEADER } from '@codebuff/common/constants/byok'
 import { isFreeMode } from '@codebuff/common/constants/free-agents'
+import { REQUESTY_BASE_URL } from '@codebuff/common/constants/requesty'
 import {
   CHATGPT_BACKEND_BASE_URL,
   CHATGPT_OAUTH_ENABLED,
@@ -24,7 +25,10 @@ import {
 
 import { WEBSITE_URL } from '../constants'
 import { getValidChatGptOAuthCredentials } from '../credentials'
-import { getByokOpenrouterApiKeyFromEnv } from '../env'
+import {
+  getByokOpenrouterApiKeyFromEnv,
+  getRequestyApiKeyFromEnv,
+} from '../env'
 import {
   createChatGptBackendFetch,
   extractChatGptAccountId,
@@ -117,6 +121,18 @@ export async function getModelForRequest(
 ): Promise<ModelResult> {
   const { apiKey, model, skipChatGptOAuth, costMode } = params
 
+  // Direct Requesty route: when REQUESTY_API_KEY is set, send chat completions
+  // straight to the Requesty OpenAI-compatible router instead of the Codebuff
+  // backend. Model ids use the same `provider/model` form (e.g.
+  // "openai/gpt-4o-mini"), so no remapping is required.
+  const requestyApiKey = getRequestyApiKeyFromEnv()
+  if (requestyApiKey) {
+    return {
+      model: createRequestyModel(model, requestyApiKey),
+      isChatGptOAuth: false,
+    }
+  }
+
   // Check if we should use ChatGPT OAuth direct
   // Only attempt for allowlisted models; non-allowlisted models silently fall through to backend.
   if (
@@ -191,6 +207,30 @@ function createOpenAIOAuthModel(
   })
 }
 
+/**
+ * Create a model that routes directly to the Requesty OpenAI-compatible router
+ * (https://router.requesty.ai/v1). Used when REQUESTY_API_KEY is set.
+ *
+ * This mirrors createOpenAIOAuthModel: it builds an OpenAICompatibleChatLanguageModel
+ * pointed at a fixed upstream base URL with a Bearer API key, rather than going
+ * through the Codebuff backend. Model ids use the same `provider/model` form as
+ * OpenRouter (e.g. "openai/gpt-4o-mini").
+ */
+function createRequestyModel(model: string, apiKey: string): LanguageModel {
+  return new OpenAICompatibleChatLanguageModel(model, {
+    provider: 'requesty',
+    url: ({ path }) => `${REQUESTY_BASE_URL}${path}`,
+    headers: () => ({
+      Authorization: `Bearer ${apiKey}`,
+      'Content-Type': 'application/json',
+      'user-agent': `ai-sdk/openai-compatible/${VERSION}/codebuff-requesty`,
+    }),
+    fetch: undefined,
+    includeUsage: undefined,
+    supportsStructuredOutputs: true,
+  })
+}
+
 /**
  * Create a model that routes through the Codebuff backend.
  * This is the existing behavior - requests go to Codebuff backend which forwards to OpenRouter.