Skip to content

Use Claude Code, Codex, VSCode, Pi, and OpenCode (and 6 other harnesses) for free (1.3B+ free tokens) from your terminal, app, IDE, or phone, and now from the browser with native browser sessions (multi-harness + multi-model) like OpenClaw (voice supported + ToS friendly)

Python

Add Vercel AI Gateway provider (#984)

Ali Khokhar020bbef64b0e
diff --git a/.env.example b/.env.example
index ef1b45c5..97f4b4ed 100644
--- a/.env.example
+++ b/.env.example
@@ -34,6 +34,10 @@ MINIMAX_API_KEY=""
OPENCODE_API_KEY=""
+# Vercel AI Gateway Config (OpenAI-compatible Chat Completions at ai-gateway.vercel.sh/v1)
+AI_GATEWAY_API_KEY=""
+
+
# Z.ai Config (Anthropic-compatible Messages at api.z.ai/api/anthropic/v1)
ZAI_API_KEY=""
@@ -73,7 +77,7 @@ OLLAMA_BASE_URL="http://localhost:11434"
# All Claude model requests are mapped to these models, plain model is fallback
# Format: provider_type/model/name
-# Valid providers: "nvidia_nim" | "open_router" | "gemini" | "deepseek" | "mistral" | "mistral_codestral" | "opencode" | "opencode_go" | "wafer" | "kimi" | "minimax" | "cerebras" | "groq" | "fireworks" | "cloudflare" | "zai" | "lmstudio" | "llamacpp" | "ollama"
+# Valid providers: "nvidia_nim" | "open_router" | "gemini" | "deepseek" | "mistral" | "mistral_codestral" | "opencode" | "opencode_go" | "vercel" | "wafer" | "kimi" | "minimax" | "cerebras" | "groq" | "fireworks" | "cloudflare" | "zai" | "lmstudio" | "llamacpp" | "ollama"
MODEL_OPUS=
MODEL_SONNET=
MODEL_HAIKU=
@@ -95,6 +99,7 @@ FCC_SMOKE_MODEL_WAFER=
FCC_SMOKE_MODEL_MINIMAX=
FCC_SMOKE_MODEL_OPENCODE=
FCC_SMOKE_MODEL_OPENCODE_GO=
+FCC_SMOKE_MODEL_VERCEL=
FCC_SMOKE_MODEL_ZAI=
FCC_SMOKE_MODEL_FIREWORKS=
FCC_SMOKE_MODEL_CLOUDFLARE=
@@ -129,6 +134,7 @@ WAFER_PROXY=""
MINIMAX_PROXY=""
OPENCODE_PROXY=""
OPENCODE_GO_PROXY=""
+VERCEL_AI_GATEWAY_PROXY=""
ZAI_PROXY=""
FIREWORKS_PROXY=""
CLOUDFLARE_PROXY=""
diff --git a/ARCHITECTURE.md b/ARCHITECTURE.md
index 3170fb5b..0873bcdb 100644
--- a/ARCHITECTURE.md
+++ b/ARCHITECTURE.md
@@ -371,6 +371,9 @@ where supported, and returning Anthropic SSE strings to the service layer.
Provider-specific inputs that do not apply to other upstreams, such as
Cloudflare's account ID, stay in that provider's factory/client instead of being
added to shared `ProviderConfig`.
+Gateway providers such as Vercel AI Gateway stay thin when their documented
+OpenAI-compatible Chat Completions behavior matches shared transport policy;
+provider-specific gateway options pass through request `extra_body`.
### Adding A Provider
diff --git a/README.md b/README.md
index bb200067..9b5c86a4 100644
--- a/README.md
+++ b/README.md
@@ -57,7 +57,7 @@ Free Claude Code routes Anthropic Messages API traffic from Claude Code (CLI and
- Drop-in proxy for Claude Code's Anthropic API calls (`/v1/messages`, `/v1/models`).
- Drop-in proxy for Codex via the OpenAI Responses API (`/v1/responses`).
- `fcc-claude` and `fcc-codex` launchers that read the current Admin UI port and auth token each time they start.
-- 19 provider backends: NVIDIA NIM, OpenRouter, Google AI Studio (Gemini), DeepSeek, Mistral La Plateforme, Mistral Codestral, OpenCode Zen, OpenCode Go, Wafer, Kimi, MiniMax, Cerebras Inference, Groq, Fireworks AI, Cloudflare, Z.ai, LM Studio, llama.cpp, and Ollama.
+- 20 provider backends: NVIDIA NIM, OpenRouter, Google AI Studio (Gemini), DeepSeek, Mistral La Plateforme, Mistral Codestral, OpenCode Zen, OpenCode Go, Vercel AI Gateway, Wafer, Kimi, MiniMax, Cerebras Inference, Groq, Fireworks AI, Cloudflare, Z.ai, LM Studio, llama.cpp, and Ollama.
- Per-model routing for Claude Code: send Opus, Sonnet, Haiku, and fallback traffic to different providers.
- Native Claude Code `/model` picker support through the proxy's `/v1/models` endpoint (see [Model Picker](#model-picker)).
- Native Codex `/model` picker support when launched through `fcc-codex`, using a generated local model catalog.
@@ -253,7 +253,17 @@ Popular examples:
Browse available models at [opencode.ai](https://opencode.ai).
-### 9. [Wafer](https://wafer.ai/)
+### 9. [Vercel AI Gateway](https://vercel.com/docs/ai-gateway)
+
+Create an AI Gateway API key from Vercel, then paste it into `AI_GATEWAY_API_KEY` in the Admin UI.
+
+Set `MODEL` to a Vercel model slug such as `vercel/openai/gpt-5.5`, `vercel/anthropic/claude-sonnet-4.6`, or `vercel/google/gemini-3.1-flash-lite-preview`.
+
+Vercel AI Gateway exposes an OpenAI-compatible endpoint at `https://ai-gateway.vercel.sh/v1`, supports `GET /models`, streaming Chat Completions, tool calls, provider options, and gateway fallbacks. FCC routes it through the shared OpenAI-chat transport and preserves request `extra_body` for Vercel-specific options.
+
+Browse models at [Vercel models and providers](https://vercel.com/docs/ai-gateway/models-and-providers).
+
+### 10. [Wafer](https://wafer.ai/)
Get a key from [wafer.ai](https://wafer.ai). In the Admin UI, paste it into `WAFER_API_KEY`, then set `MODEL` to a Wafer Pass model such as `wafer/DeepSeek-V4-Pro`.
@@ -266,7 +276,7 @@ Popular examples:
This provider uses Wafer's Anthropic-compatible endpoint at `https://pass.wafer.ai/v1/messages`.
-### 10. [Kimi](https://platform.moonshot.ai/)
+### 11. [Kimi](https://platform.moonshot.ai/)
Get a key at [platform.moonshot.ai/console/api-keys](https://platform.moonshot.ai/console/api-keys).
@@ -276,7 +286,7 @@ This provider calls Kimi's **Anthropic-compatible** Messages API (`https://api.m
Browse models at [platform.moonshot.ai](https://platform.moonshot.ai).
-### 11. [MiniMax](https://platform.minimax.io/)
+### 12. [MiniMax](https://platform.minimax.io/)
Get a key from [MiniMax](https://platform.minimax.io/user-center/basic-information/interface-key).
@@ -284,7 +294,7 @@ In the Admin UI, paste it into `MINIMAX_API_KEY`, then set `MODEL` to a MiniMax
This provider calls MiniMax's **Anthropic-compatible** Messages API (`https://api.minimax.io/anthropic/v1/messages`). `MiniMax-M3` is the recommended default because MiniMax documents controllable Anthropic thinking for that model; other MiniMax models remain discoverable through the provider model list.
-### 12. [Cerebras Inference](https://inference-docs.cerebras.ai/quickstart)
+### 13. [Cerebras Inference](https://inference-docs.cerebras.ai/quickstart)
Sign up and create an API key in the [Cerebras Cloud Console](https://cloud.cerebras.ai) (see [Quickstart](https://inference-docs.cerebras.ai/quickstart)).
@@ -292,7 +302,7 @@ In the Admin UI, set `CEREBRAS_API_KEY`, then route with `MODEL` such as `cerebr
Cerebras exposes an OpenAI-compatible API at `https://api.cerebras.ai/v1` ([OpenAI compatibility](https://inference-docs.cerebras.ai/resources/openai)). Non-standard request fields should go in `extra_body` when using the OpenAI client; see the same page. For reasoning models and parameters, see [Reasoning](https://inference-docs.cerebras.ai/capabilities/reasoning). This proxy follows other OpenAI-compat adapters for thinking via `reasoning_content` when Claude-style thinking is enabled.
-### 13. [Groq](https://console.groq.com/)
+### 14. [Groq](https://console.groq.com/)
Get an API key at [console.groq.com/keys](https://console.groq.com/keys).
@@ -304,7 +314,7 @@ Reasoning-heavy models expose extra knobs documented under [Groq reasoning](http
Browse models at [console.groq.com/docs/models](https://console.groq.com/docs/models).
-### 14. [Fireworks AI](https://fireworks.ai/)
+### 15. [Fireworks AI](https://fireworks.ai/)
Get an API key at [fireworks.ai/account/api-keys](https://fireworks.ai/account/api-keys).
@@ -314,7 +324,7 @@ Fireworks exposes an **Anthropic-compatible** Messages API at `https://api.firew
Browse models at [fireworks.ai/models](https://fireworks.ai/models).
-### 15. [Cloudflare](https://developers.cloudflare.com/workers-ai/)
+### 16. [Cloudflare](https://developers.cloudflare.com/workers-ai/)
Create a Cloudflare API token and copy your account ID from the Cloudflare dashboard.
@@ -322,7 +332,7 @@ In the Admin UI, set `CLOUDFLARE_API_TOKEN` and `CLOUDFLARE_ACCOUNT_ID`, then se
This provider calls Cloudflare's account-scoped **OpenAI-compatible** Chat Completions API at `https://api.cloudflare.com/client/v4/accounts/<account_id>/ai/v1/chat/completions`. Use literal Workers AI model IDs, including the `@cf/` prefix when the catalog model includes it.
-### 16. [Z.ai](https://z.ai/)
+### 17. [Z.ai](https://z.ai/)
Get an API key at [Z.ai/manage-apikey/apikey-list](https://z.ai/manage-apikey/apikey-list).
@@ -337,13 +347,13 @@ Popular examples:
Browse models at [Z.ai](https://z.ai).
-### 17. [LM Studio](https://lmstudio.ai/)
+### 18. [LM Studio](https://lmstudio.ai/)
Start LM Studio's local server and load a model. In the Admin UI, keep or update `LM_STUDIO_BASE_URL`, then set `MODEL` to the model identifier shown by LM Studio, prefixed with `lmstudio/`.
Prefer models with tool-use support for Claude Code workflows.
-### 18. [llama.cpp](https://github.com/ggml-org/llama.cpp)
+### 19. [llama.cpp](https://github.com/ggml-org/llama.cpp)
Start `llama-server` with an Anthropic-compatible `/v1/messages` endpoint and enough context for Claude Code requests.
@@ -351,7 +361,7 @@ In the Admin UI, keep or update `LLAMACPP_BASE_URL`, then set `MODEL` to the loc
For local coding models, context size matters. If llama.cpp returns HTTP 400 for normal Claude Code requests, increase `--ctx-size` and verify the model/server build supports the requested features.
-### 19. [Ollama](https://ollama.com/)
+### 20. [Ollama](https://ollama.com/)
Run Ollama and pull a model:
@@ -364,7 +374,7 @@ In the Admin UI, keep or update `OLLAMA_BASE_URL`, then set `MODEL` to the same
`OLLAMA_BASE_URL` is the Ollama server root; do not append `/v1`. Example model slugs include `ollama/llama3.1` and `ollama/llama3.1:8b`.
-### 20. Mix Providers By Model Tier
+### 21. Mix Providers By Model Tier
Each model tier can use a different provider by setting `MODEL_OPUS`, `MODEL_SONNET`, and `MODEL_HAIKU` in the Admin UI. Leave a tier blank to inherit `MODEL`. These tier overrides apply to Claude model names that contain `opus`, `sonnet`, or `haiku`. Codex uses the Admin `MODEL` default through `fcc-codex` unless a session requests a provider-prefixed slug directly.
@@ -578,7 +588,7 @@ Important pieces:
- Responses requests convert to Anthropic Messages internally, then share the same model router, normalizer, and provider adapters.
- `fcc-codex` registers a custom `fcc` provider that points Codex at the local proxy's `/v1/responses` endpoint.
- Model routing resolves Claude model names to `MODEL_OPUS`, `MODEL_SONNET`, `MODEL_HAIKU`, or `MODEL`.
-- NIM, DeepSeek, OpenCode Zen, and OpenCode Go use OpenAI chat streaming translated into Anthropic SSE.
+- NIM, Gemini, DeepSeek, Mistral, Codestral, OpenCode Zen, OpenCode Go, Vercel AI Gateway, Cerebras, Groq, and Cloudflare use OpenAI chat streaming translated into Anthropic SSE.
- Wafer, OpenRouter, Kimi, MiniMax, Fireworks AI, Z.ai, LM Studio, llama.cpp, and Ollama use Anthropic Messages style transports where applicable (with provider-specific quirks and model-list URLs).
- The proxy normalizes thinking blocks, tool calls, token usage metadata, and provider errors into the shape each client expects.
- Request optimizations answer trivial Claude Code probes locally to save latency and quota.
diff --git a/api/admin_config/provider_manifest.py b/api/admin_config/provider_manifest.py
index 65f5127f..a88602a4 100644
--- a/api/admin_config/provider_manifest.py
+++ b/api/admin_config/provider_manifest.py
@@ -30,6 +30,13 @@ _PROVIDER_FIELD_OVERRIDES: dict[str, dict[str, Any]] = {
"gateway (opencode.ai/zen/go/v1); single key from opencode.ai/auth."
),
},
+ "AI_GATEWAY_API_KEY": {
+ "label": "Vercel AI Gateway API Key",
+ "description": (
+ "Vercel AI Gateway API key for the OpenAI-compatible endpoint at "
+ "ai-gateway.vercel.sh/v1."
+ ),
+ },
"ZAI_API_KEY": {
"label": "Z.ai API Key",
"description": "Z.ai Coding Plan API key.",
diff --git a/config/provider_catalog.py b/config/provider_catalog.py
index e2496861..5895c057 100644
--- a/config/provider_catalog.py
+++ b/config/provider_catalog.py
@@ -29,6 +29,7 @@ LLAMACPP_DEFAULT_BASE = "http://localhost:8080/v1"
OLLAMA_DEFAULT_BASE = "http://localhost:11434"
OPENCODE_DEFAULT_BASE = "https://opencode.ai/zen/v1"
OPENCODE_GO_DEFAULT_BASE = "https://opencode.ai/zen/go/v1"
+VERCEL_AI_GATEWAY_DEFAULT_BASE = "https://ai-gateway.vercel.sh/v1"
# Z.ai Anthropic-compatible Messages API (not OpenAI Coding Plan chat completions).
ZAI_DEFAULT_BASE = "https://api.z.ai/api/anthropic/v1"
# Google AI Studio Gemini API OpenAI-compat layer (not Vertex AI).
@@ -142,6 +143,17 @@ PROVIDER_CATALOG: dict[str, ProviderDescriptor] = {
proxy_attr="opencode_go_proxy",
capabilities=("chat", "streaming", "tools", "thinking", "rate_limit"),
),
+ "vercel": ProviderDescriptor(
+ provider_id="vercel",
+ display_name="Vercel AI Gateway",
+ transport_type="openai_chat",
+ credential_env="AI_GATEWAY_API_KEY",
+ credential_url="https://vercel.com/docs/ai-gateway",
+ credential_attr="vercel_ai_gateway_api_key",
+ default_base_url=VERCEL_AI_GATEWAY_DEFAULT_BASE,
+ proxy_attr="vercel_ai_gateway_proxy",
+ capabilities=("chat", "streaming", "tools", "thinking", "rate_limit"),
+ ),
"wafer": ProviderDescriptor(
provider_id="wafer",
display_name="Wafer",
@@ -301,10 +313,10 @@ PROVIDER_CATALOG: dict[str, ProviderDescriptor] = {
}
# Key order:
-# NVIDIA NIM first (README default), DeepSeek fourth, Wafer ninth / Kimi tenth;
-# MiniMax follows the native Anthropic remotes; then cerebras / groq / fireworks /
-# Cloudflare overlap; remainder and locals last per project plan (
-# github.com/cheahjs/free-llm-api-resources Free Providers TOC as rough guide beyond fixed slots).
+# NVIDIA NIM first (README default), DeepSeek fourth, OpenCode gateways adjacent,
+# Vercel follows gateway-style remotes, then native Anthropic remotes and locals
+# per project plan (github.com/cheahjs/free-llm-api-resources Free Providers TOC
+# as rough guide beyond fixed slots).
# ``SUPPORTED_PROVIDER_IDS`` inherits this insertion order for UI and error-message listing.
SUPPORTED_PROVIDER_IDS: tuple[str, ...] = tuple(PROVIDER_CATALOG.keys())
diff --git a/config/settings.py b/config/settings.py
index a50eb448..d1ed657e 100644
--- a/config/settings.py
+++ b/config/settings.py
@@ -44,6 +44,11 @@ class Settings(BaseSettings):
# Same key from opencode.ai/auth; zen uses prefix ``opencode/``, Go uses ``opencode_go/``.
opencode_api_key: str = Field(default="", validation_alias="OPENCODE_API_KEY")
+ # ==================== Vercel AI Gateway ====================
+ vercel_ai_gateway_api_key: str = Field(
+ default="", validation_alias="AI_GATEWAY_API_KEY"
+ )
+
# ==================== Z.ai Config ====================
zai_api_key: str = Field(default="", validation_alias="ZAI_API_KEY")
@@ -123,6 +128,9 @@ class Settings(BaseSettings):
minimax_proxy: str = Field(default="", validation_alias="MINIMAX_PROXY")
opencode_proxy: str = Field(default="", validation_alias="OPENCODE_PROXY")
opencode_go_proxy: str = Field(default="", validation_alias="OPENCODE_GO_PROXY")
+ vercel_ai_gateway_proxy: str = Field(
+ default="", validation_alias="VERCEL_AI_GATEWAY_PROXY"
+ )
zai_proxy: str = Field(default="", validation_alias="ZAI_PROXY")
fireworks_proxy: str = Field(default="", validation_alias="FIREWORKS_PROXY")
cloudflare_proxy: str = Field(default="", validation_alias="CLOUDFLARE_PROXY")
diff --git a/providers/defaults.py b/providers/defaults.py
index da566a7b..df469c4a 100644
--- a/providers/defaults.py
+++ b/providers/defaults.py
@@ -17,6 +17,7 @@ from config.provider_catalog import (
OPENCODE_DEFAULT_BASE,
OPENCODE_GO_DEFAULT_BASE,
OPENROUTER_DEFAULT_BASE,
+ VERCEL_AI_GATEWAY_DEFAULT_BASE,
WAFER_DEFAULT_BASE,
ZAI_DEFAULT_BASE,
)
@@ -38,6 +39,7 @@ __all__ = (
"OPENCODE_DEFAULT_BASE",
"OPENCODE_GO_DEFAULT_BASE",
"OPENROUTER_DEFAULT_BASE",
+ "VERCEL_AI_GATEWAY_DEFAULT_BASE",
"WAFER_DEFAULT_BASE",
"ZAI_DEFAULT_BASE",
)
diff --git a/providers/runtime/factory.py b/providers/runtime/factory.py
index 442e32b1..2b013672 100644
--- a/providers/runtime/factory.py
+++ b/providers/runtime/factory.py
@@ -95,6 +95,12 @@ def _create_opencode_go(config: ProviderConfig, _settings: Settings) -> BaseProv
return OpenCodeProvider(config, provider_name="OPENCODE_GO")
+def _create_vercel(config: ProviderConfig, _settings: Settings) -> BaseProvider:
+ from providers.vercel import VercelProvider
+
+ return VercelProvider(config)
+
+
def _create_zai(config: ProviderConfig, _settings: Settings) -> BaseProvider:
from providers.zai import ZaiProvider
@@ -140,6 +146,7 @@ PROVIDER_FACTORIES: dict[str, ProviderFactory] = {
"mistral_codestral": _create_mistral_codestral,
"opencode": _create_opencode,
"opencode_go": _create_opencode_go,
+ "vercel": _create_vercel,
"wafer": _create_wafer,
"kimi": _create_kimi,
"minimax": _create_minimax,
diff --git a/providers/vercel/__init__.py b/providers/vercel/__init__.py
new file mode 100644
index 00000000..43bfac74
--- /dev/null
+++ b/providers/vercel/__init__.py
@@ -0,0 +1,7 @@
+"""Vercel AI Gateway OpenAI-compatible adapter."""
+
+from providers.defaults import VERCEL_AI_GATEWAY_DEFAULT_BASE
+
+from .client import VercelProvider
+
+__all__ = ["VERCEL_AI_GATEWAY_DEFAULT_BASE", "VercelProvider"]
diff --git a/providers/vercel/client.py b/providers/vercel/client.py
new file mode 100644
index 00000000..d7183330
--- /dev/null
+++ b/providers/vercel/client.py
@@ -0,0 +1,37 @@
+"""Vercel AI Gateway provider implementation."""
+
+from typing import Any
+
+from providers.base import ProviderConfig
+from providers.defaults import VERCEL_AI_GATEWAY_DEFAULT_BASE
+from providers.transports.openai_chat import (
+ OpenAIChatRequestPolicy,
+ OpenAIChatTransport,
+ build_openai_chat_request_body,
+)
+
+_REQUEST_POLICY = OpenAIChatRequestPolicy(
+ provider_name="VERCEL",
+ include_extra_body=True,
+)
+
+
+class VercelProvider(OpenAIChatTransport):
+ """Vercel AI Gateway at ``https://ai-gateway.vercel.sh/v1``."""
+
+ def __init__(self, config: ProviderConfig):
+ super().__init__(
+ config,
+ provider_name="VERCEL",
+ base_url=config.base_url or VERCEL_AI_GATEWAY_DEFAULT_BASE,
+ api_key=config.api_key,
+ )
+
+ def _build_request_body(
+ self, request: Any, thinking_enabled: bool | None = None
+ ) -> dict:
+ return build_openai_chat_request_body(
+ request,
+ thinking_enabled=self._is_thinking_enabled(request, thinking_enabled),
+ policy=_REQUEST_POLICY,
+ )
diff --git a/pyproject.toml b/pyproject.toml
index 664adddd..594e04fa 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
[project]
name = "free-claude-code"
-version = "2.5.4"
+version = "2.6.0"
description = "Middleware between Claude Code CLI (Anthropic API) and NVIDIA NIM"
readme = "README.md"
requires-python = ">=3.14.0"
diff --git a/smoke/lib/config.py b/smoke/lib/config.py
index 0d55425b..05a06df8 100644
--- a/smoke/lib/config.py
+++ b/smoke/lib/config.py
@@ -53,6 +53,7 @@ PROVIDER_SMOKE_DEFAULT_MODELS: dict[str, str] = {
"minimax": "minimax/MiniMax-M3",
"opencode": "opencode/gpt-5.3-codex",
"opencode_go": "opencode_go/minimax-m2.7",
+ "vercel": "vercel/openai/gpt-5.5",
"zai": "zai/glm-5.1",
"gemini": "gemini/models/gemini-3.1-flash-lite",
"groq": "groq/llama-3.3-70b-versatile",
@@ -252,6 +253,8 @@ class SmokeConfig:
return bool(self.settings.opencode_api_key.strip())
if provider == "opencode_go":
return bool(self.settings.opencode_api_key.strip())
+ if provider == "vercel":
+ return bool(self.settings.vercel_ai_gateway_api_key.strip())
if provider == "zai":
return bool(self.settings.zai_api_key.strip())
if provider == "gemini":
diff --git a/tests/config/test_config.py b/tests/config/test_config.py
index 60d9835b..b3a03780 100644
--- a/tests/config/test_config.py
+++ b/tests/config/test_config.py
@@ -306,6 +306,16 @@ class TestSettings:
assert settings.cloudflare_account_id == "cf-account"
assert settings.cloudflare_proxy == "http://proxy.test:8080"
+ def test_vercel_settings_from_env(self, monkeypatch):
+ """Vercel AI Gateway key and proxy env vars load into settings."""
+ from config.settings import Settings
+
+ monkeypatch.setenv("AI_GATEWAY_API_KEY", "vercel-key")
+ monkeypatch.setenv("VERCEL_AI_GATEWAY_PROXY", "http://proxy.test:8080")
+ settings = Settings()
+ assert settings.vercel_ai_gateway_api_key == "vercel-key"
+ assert settings.vercel_ai_gateway_proxy == "http://proxy.test:8080"
+
def test_per_model_thinking_from_env(self, monkeypatch):
"""Per-model thinking env vars are loaded into settings."""
from config.settings import Settings
@@ -883,6 +893,7 @@ class TestPerModelMapping:
assert (
parse_provider_type("cloudflare/@cf/moonshotai/kimi-k2.6") == "cloudflare"
)
+ assert parse_provider_type("vercel/openai/gpt-5.5") == "vercel"
assert parse_provider_type("gemini/models/gemini-3.1-flash-lite") == "gemini"
assert parse_provider_type("groq/llama-3.3-70b-versatile") == "groq"
assert parse_provider_type("cerebras/llama3.1-8b") == "cerebras"
@@ -907,6 +918,7 @@ class TestPerModelMapping:
parse_model_name("cloudflare/@cf/moonshotai/kimi-k2.6")
== "@cf/moonshotai/kimi-k2.6"
)
+ assert parse_model_name("vercel/openai/gpt-5.5") == "openai/gpt-5.5"
assert (
parse_model_name("gemini/models/gemini-3.1-flash-lite")
== "models/gemini-3.1-flash-lite"
diff --git a/tests/contracts/test_feature_manifest.py b/tests/contracts/test_feature_manifest.py
index fb0fc897..f426c373 100644
--- a/tests/contracts/test_feature_manifest.py
+++ b/tests/contracts/test_feature_manifest.py
@@ -19,6 +19,7 @@ from providers.nvidia_nim import NvidiaNimProvider
from providers.ollama import OllamaProvider
from providers.open_router import OpenRouterProvider
from providers.opencode import OpenCodeProvider
+from providers.vercel import VercelProvider
from providers.wafer import WaferProvider
from providers.zai import ZaiProvider
from smoke.features import FEATURE_INVENTORY, README_FEATURES, feature_ids
@@ -92,6 +93,7 @@ def test_provider_and_platform_registries_include_advertised_builtins() -> None:
"wafer": WaferProvider,
"opencode": OpenCodeProvider,
"opencode_go": OpenCodeProvider,
+ "vercel": VercelProvider,
"zai": ZaiProvider,
"gemini": GeminiProvider,
"groq": GroqProvider,
diff --git a/tests/contracts/test_provider_catalog_order.py b/tests/contracts/test_provider_catalog_order.py
index a8b6fca7..185ebe33 100644
--- a/tests/contracts/test_provider_catalog_order.py
+++ b/tests/contracts/test_provider_catalog_order.py
@@ -11,6 +11,7 @@ _EXPECTED_PROVIDER_ORDER: tuple[str, ...] = (
"mistral_codestral",
"opencode",
"opencode_go",
+ "vercel",
"wafer",
"kimi",
"minimax",
@@ -26,7 +27,7 @@ _EXPECTED_PROVIDER_ORDER: tuple[str, ...] = (
def test_provider_catalog_key_order_matches_canonical_plan() -> None:
- """NIM first; DeepSeek fourth; Wafer ninth / Kimi tenth (see contributor plan)."""
+ """NIM first; OpenCode pair stays adjacent; Vercel follows gateway providers."""
assert tuple(PROVIDER_CATALOG.keys()) == _EXPECTED_PROVIDER_ORDER
assert SUPPORTED_PROVIDER_IDS == _EXPECTED_PROVIDER_ORDER
diff --git a/tests/contracts/test_smoke_config.py b/tests/contracts/test_smoke_config.py
index 035a30fb..e569f983 100644
--- a/tests/contracts/test_smoke_config.py
+++ b/tests/contracts/test_smoke_config.py
@@ -35,6 +35,7 @@ def _settings(**overrides):
"wafer_api_key": "",
"minimax_api_key": "",
"opencode_api_key": "",
+ "vercel_ai_gateway_api_key": "",
"zai_api_key": "",
"gemini_api_key": "",
"groq_api_key": "",
@@ -192,6 +193,22 @@ def test_cloudflare_provider_configuration_missing_account_is_unconfigured() ->
assert not config.has_provider_configuration("cloudflare")
+def test_vercel_provider_configuration_uses_api_key(monkeypatch) -> None:
+ monkeypatch.delenv("FCC_SMOKE_MODEL_VERCEL", raising=False)
+ config = _smoke_config(
+ settings=_settings(
+ model="ollama/llama3.1",
+ ollama_base_url="",
+ vercel_ai_gateway_api_key="vercel-key",
+ )
+ )
+
+ assert config.has_provider_configuration("vercel")
+ models = config.provider_smoke_models()
+ assert models[0].provider == "vercel"
+ assert models[0].full_model == PROVIDER_SMOKE_DEFAULT_MODELS["vercel"]
+
+
def test_provider_smoke_model_override_accepts_model_name_without_prefix(
monkeypatch,
) -> None:
diff --git a/tests/providers/test_provider_runtime.py b/tests/providers/test_provider_runtime.py
index b744cf1e..66691b34 100644
--- a/tests/providers/test_provider_runtime.py
+++ b/tests/providers/test_provider_runtime.py
@@ -29,6 +29,7 @@ from providers.ollama import OllamaProvider
from providers.open_router import OpenRouterProvider
from providers.opencode import OpenCodeProvider
from providers.runtime import ProviderRuntime, build_provider_config, create_provider
+from providers.vercel import VERCEL_AI_GATEWAY_DEFAULT_BASE, VercelProvider
from providers.wafer import WaferProvider
from providers.zai import ZaiProvider
@@ -47,6 +48,7 @@ def _make_settings(**overrides):
mock.wafer_api_key = "test_wafer_key"
mock.minimax_api_key = "test_minimax_key"
mock.opencode_api_key = "test_opencode_key"
+ mock.vercel_ai_gateway_api_key = "test_vercel_key"
mock.zai_api_key = "test_zai_key"
mock.lm_studio_base_url = "http://localhost:1234/v1"
mock.llamacpp_base_url = "http://localhost:8080/v1"
@@ -63,6 +65,7 @@ def _make_settings(**overrides):
mock.minimax_proxy = ""
mock.opencode_proxy = ""
mock.opencode_go_proxy = ""
+ mock.vercel_ai_gateway_proxy = ""
mock.zai_proxy = ""
mock.fireworks_proxy = ""
mock.fireworks_api_key = "test_fireworks_key"
@@ -200,6 +203,29 @@ def test_build_provider_config_opencode_go_uses_opencode_api_key() -> None:
assert config.api_key == "shared-opencode-token"
+def test_vercel_descriptor_uses_openai_chat_gateway() -> None:
+ descriptor = PROVIDER_CATALOG["vercel"]
+
+ assert descriptor.transport_type == "openai_chat"
+ assert descriptor.default_base_url == VERCEL_AI_GATEWAY_DEFAULT_BASE
+ assert descriptor.credential_env == "AI_GATEWAY_API_KEY"
+ assert descriptor.proxy_attr == "vercel_ai_gateway_proxy"
+ assert "thinking" in descriptor.capabilities
+
+
+def test_build_provider_config_vercel_uses_gateway_key_and_proxy() -> None:
+ descriptor = PROVIDER_CATALOG["vercel"]
+ settings = _make_settings(
+ vercel_ai_gateway_api_key="vercel-token",
+ vercel_ai_gateway_proxy="http://proxy.test:8080",
+ )
+
+ config = build_provider_config(descriptor, settings)
+
+ assert config.api_key == "vercel-token"
+ assert config.proxy == "http://proxy.test:8080"
+
+
def test_create_provider_uses_native_openrouter_by_default():
with patch("httpx.AsyncClient"):
provider = create_provider("open_router", _make_settings())
@@ -215,6 +241,7 @@ def test_create_provider_instantiates_each_builtin():
fireworks_api_key="test_fireworks_key",
cloudflare_api_token="test_cloudflare_token",
cloudflare_account_id="test_cloudflare_account",
+ vercel_ai_gateway_api_key="test_vercel_key",
kimi_api_key="test_kimi_key",
)
cases = {
@@ -232,6 +259,7 @@ def test_create_provider_instantiates_each_builtin():
"wafer": WaferProvider,
"opencode": OpenCodeProvider,
"opencode_go": OpenCodeProvider,
+ "vercel": VercelProvider,
"zai": ZaiProvider,
"gemini": GeminiProvider,
"groq": GroqProvider,
diff --git a/tests/providers/test_vercel.py b/tests/providers/test_vercel.py
new file mode 100644
index 00000000..9269e3d3
--- /dev/null
+++ b/tests/providers/test_vercel.py
@@ -0,0 +1,187 @@
+"""Tests for Vercel AI Gateway provider."""
+
+from contextlib import asynccontextmanager
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+
+from providers.base import ProviderConfig
+from providers.vercel import VERCEL_AI_GATEWAY_DEFAULT_BASE, VercelProvider
+
+
+class MockMessage:
+ def __init__(self, role, content):
+ self.role = role
+ self.content = content
+
+
+class MockRequest:
+ def __init__(self, **kwargs):
+ self.model = "openai/gpt-5.5"
+ self.messages = [MockMessage("user", "Hello")]
+ self.max_tokens = 100
+ self.temperature = 0.5
+ self.top_p = 0.9
+ self.system = "System prompt"
+ self.stop_sequences = None
+ self.tools = []
+ self.thinking = MagicMock()
+ self.thinking.enabled = True
+ for key, value in kwargs.items():
+ setattr(self, key, value)
+
+
+def vercel_config():
+ return ProviderConfig(
+ api_key="test_vercel_key",
+ base_url=VERCEL_AI_GATEWAY_DEFAULT_BASE,
+ rate_limit=10,
+ rate_window=60,
+ enable_thinking=True,
+ )
+
+
[email protected](autouse=True)
+def mock_rate_limiter():
+ @asynccontextmanager
+ async def _slot():
+ yield
+
+ with patch("providers.transports.openai_chat.transport.GlobalRateLimiter") as mock:
+ instance = mock.get_scoped_instance.return_value
+
+ async def _passthrough(fn, *args, **kwargs):
+ return await fn(*args, **kwargs)
+
+ instance.execute_with_retry = AsyncMock(side_effect=_passthrough)
+ instance.concurrency_slot.side_effect = _slot
+ yield instance
+
+
+def vercel_provider(vercel_config):
+ return VercelProvider(vercel_config)
+
+
+def test_default_base_url_constant():
+ assert VERCEL_AI_GATEWAY_DEFAULT_BASE == "https://ai-gateway.vercel.sh/v1"
+
+
+def test_init_uses_default_base_url_and_api_key(vercel_config):
+ with patch("providers.transports.openai_chat.transport.AsyncOpenAI") as mock_openai:
+ provider = VercelProvider(vercel_config)
+
+ assert provider._api_key == "test_vercel_key"
+ assert provider._base_url == VERCEL_AI_GATEWAY_DEFAULT_BASE
+ mock_openai.assert_called_once()
+
+
+def test_init_strips_trailing_slash(vercel_config):
+ config = vercel_config.model_copy(
+ update={"base_url": f"{VERCEL_AI_GATEWAY_DEFAULT_BASE}/"}
+ )
+
+ with patch("providers.transports.openai_chat.transport.AsyncOpenAI"):
+ provider = VercelProvider(config)
+
+ assert provider._base_url == VERCEL_AI_GATEWAY_DEFAULT_BASE
+
+
+def test_build_request_body_keeps_max_tokens(vercel_provider):
+ with patch(
+ "providers.transports.openai_chat.request_policy.build_base_request_body"
+ ) as mock_convert:
+ mock_convert.return_value = {
+ "model": "openai/gpt-5.5",
+ "messages": [{"role": "user", "name": "alice", "content": "hi"}],
+ "max_tokens": 42,
+ }
+
+ body = vercel_provider._build_request_body(MockRequest())
+
+ assert body["messages"][0].get("name") == "alice"
+ assert body["max_tokens"] == 42
+ assert "max_completion_tokens" not in body
+
+
+def test_build_request_body_preserves_caller_extra_body(vercel_provider):
+ req = MockRequest(extra_body={"providerOptions": {"openai": {"reasoning": "low"}}})
+
+ body = vercel_provider._build_request_body(req)
+
+ assert body["extra_body"] == {"providerOptions": {"openai": {"reasoning": "low"}}}
+
+
+async def test_stream_response_text(vercel_provider):
+ mock_chunk = MagicMock()
+ mock_chunk.choices = [
+ MagicMock(
+ delta=MagicMock(
+ content="Hello from Vercel",
+ reasoning_content=None,
+ tool_calls=None,
+ ),
+ finish_reason="stop",
+ )
+ ]
+ mock_chunk.usage = MagicMock(completion_tokens=5, prompt_tokens=10)
+
+ async def mock_stream():
+ yield mock_chunk
+
+ with patch.object(
+ vercel_provider._client.chat.completions, "create", new_callable=AsyncMock
+ ) as mock_create:
+ mock_create.return_value = mock_stream()
+
+ events = [
+ event async for event in vercel_provider.stream_response(MockRequest())
+ ]
+
+ assert any(
+ '"text_delta"' in event and "Hello from Vercel" in event for event in events
+ )
+
+
+async def test_stream_response_reasoning_content(vercel_provider):
+ mock_chunk = MagicMock()
+ mock_chunk.choices = [
+ MagicMock(
+ delta=MagicMock(
+ content=None,
+ reasoning_content="Thinking via gateway",
+ tool_calls=None,
+ ),
+ finish_reason="stop",
+ )
+ ]
+ mock_chunk.usage = MagicMock(completion_tokens=2, prompt_tokens=10)
+
+ async def mock_stream():
+ yield mock_chunk
+
+ with patch.object(
+ vercel_provider._client.chat.completions, "create", new_callable=AsyncMock
+ ) as mock_create:
+ mock_create.return_value = mock_stream()
+
+ events = [
+ event async for event in vercel_provider.stream_response(MockRequest())
+ ]
+
+ assert any(
+ '"thinking_delta"' in event and "Thinking via gateway" in event
+ for event in events
+ )
+
+
+async def test_cleanup(vercel_provider):
+ vercel_provider._client = AsyncMock()
+
+ await vercel_provider.cleanup()
+
+ vercel_provider._client.close.assert_called_once()
diff --git a/uv.lock b/uv.lock
index b7734afa..b5614e93 100644
--- a/uv.lock
+++ b/uv.lock
@@ -561,7 +561,7 @@ wheels = [
[[package]]
name = "free-claude-code"
-version = "2.5.4"
+version = "2.6.0"
source = { editable = "." }
dependencies = [
{ name = "aiohttp" },