feat(providers): add Inworld as an OpenAI-compatible provider

Registers Inworld (https://api.inworld.ai/v1) as a JSON-configured
OpenAI-compatible provider: chat completions (incl. streaming usage),
the /v1/messages bridge, text-to-speech (inworld-tts-2) and speech-to-text
(inworld-stt-1).

Also fixes speech() for OpenAI-compatible providers so the provider key
resolved by get_llm_provider (e.g. INWORLD_API_KEY) is used instead of
falling through to OPENAI_API_KEY. transcription() already did this.

litellm/constants.py

@@ -1023,6 +1023,7 @@ openai_compatible_endpoints: Final[list[str]] = [
     "https://api.scx.ai/v1",
     "https://api.prisminference.com/v1",
     "https://api.reka.ai/v1",
+    "https://api.inworld.ai/v1",
     "https://gigachat.devices.sberbank.ru/api/v1",
 ]
 
@@ -1101,6 +1102,7 @@ openai_compatible_providers: Final[list] = [
     "prism",
     "sail",
     "reka",
+    "inworld",
 ]
 
 OPENAI_AUDIO_TRANSCRIPTION_PROVIDERS: Final = frozenset({"openai"} | frozenset(openai_compatible_providers))

litellm/llms/openai_like/providers.json

@@ -239,5 +239,10 @@
       "/v1/responses",
       "/v1/messages"
     ]
+  },
+  "inworld": {
+    "base_url": "https://api.inworld.ai/v1",
+    "api_key_env": "INWORLD_API_KEY",
+    "api_base_env": "INWORLD_API_BASE"
   }
 }

litellm/main.py

@@ -8465,7 +8465,8 @@ def speech(
         # set API KEY
         api_key = (
             api_key
-            or litellm.api_key  # for deepinfra/perplexity/anyscale we check in get_llm_provider and pass in the api key from there
+            or dynamic_api_key  # key resolved by get_llm_provider, e.g. INWORLD_API_KEY for JSON-configured providers
+            or litellm.api_key
             or litellm.openai_key
             or get_secret("OPENAI_API_KEY")
         )

litellm/provider_endpoints_support_backup.json

@@ -2154,6 +2154,24 @@
         "interactions": false
       }
     },
+    "inworld": {
+      "display_name": "Inworld (`inworld`)",
+      "url": "https://docs.litellm.ai/docs/providers/inworld",
+      "endpoints": {
+        "chat_completions": true,
+        "messages": true,
+        "responses": false,
+        "embeddings": false,
+        "image_generations": false,
+        "audio_transcriptions": true,
+        "audio_speech": true,
+        "moderations": false,
+        "batches": false,
+        "rerank": false,
+        "a2a": false,
+        "interactions": false
+      }
+    },
     "scaleway": {
       "display_name": "Scaleway (`scaleway`)",
       "url": "https://docs.litellm.ai/docs/providers/scaleway",

litellm/proxy/public_endpoints/provider_create_fields.json

@@ -3231,6 +3231,34 @@
     ],
     "default_model_placeholder": "reka/reka-flash"
   },
+  {
+    "provider": "INWORLD",
+    "provider_display_name": "Inworld",
+    "litellm_provider": "inworld",
+    "credential_fields": [
+      {
+        "key": "api_base",
+        "label": "API Base",
+        "placeholder": "https://api.inworld.ai/v1",
+        "tooltip": null,
+        "required": false,
+        "field_type": "text",
+        "options": null,
+        "default_value": null
+      },
+      {
+        "key": "api_key",
+        "label": "API Key",
+        "placeholder": null,
+        "tooltip": null,
+        "required": true,
+        "field_type": "password",
+        "options": null,
+        "default_value": null
+      }
+    ],
+    "default_model_placeholder": "inworld/inworld/models/gemma-4-31b-it"
+  },
   {
     "provider": "Sail",
     "provider_display_name": "Sail",

litellm/types/utils.py

@@ -4389,6 +4389,7 @@ class LlmProviders(str, Enum):
     META = "meta"
     SAIL = "sail"
     REKA = "reka"
+    INWORLD = "inworld"
     LITELLM_AGENT = "litellm_agent"
     CURSOR = "cursor"
     BEDROCK_MANTLE = "bedrock_mantle"

provider_endpoints_support.json

@@ -3214,6 +3214,24 @@
         "interactions": false
       }
     },
+    "inworld": {
+      "display_name": "Inworld (`inworld`)",
+      "url": "https://docs.litellm.ai/docs/providers/inworld",
+      "endpoints": {
+        "chat_completions": true,
+        "messages": true,
+        "responses": false,
+        "embeddings": false,
+        "image_generations": false,
+        "audio_transcriptions": true,
+        "audio_speech": true,
+        "moderations": false,
+        "batches": false,
+        "rerank": false,
+        "a2a": false,
+        "interactions": false
+      }
+    },
     "charity_engine": {
       "display_name": "Charity Engine (`charity_engine`)",
       "url": "https://docs.litellm.ai/docs/providers/charity_engine",

tests/unit/llms/openai_like/test_inworld_provider.py

@@ -0,0 +1,234 @@
+import json
+from pathlib import Path
+from typing import Final
+
+import pytest
+import respx
+
+import litellm
+from litellm.caching.llm_caching_handler import LLMClientCache
+
+_CHAT_COMPLETION: Final = {
+    "id": "chatcmpl_inworld",
+    "object": "chat.completion",
+    "created": 1_790_000_000,
+    "model": "inworld/gemma-4-31b-it",
+    "choices": [
+        {
+            "index": 0,
+            "message": {"role": "assistant", "content": "Hello from Inworld"},
+            "finish_reason": "stop",
+        }
+    ],
+    "usage": {"prompt_tokens": 4, "completion_tokens": 3, "total_tokens": 7},
+}
+
+
+def test_inworld_is_a_registered_provider():
+    assert litellm.LlmProviders.INWORLD.value == "inworld"
+    assert "inworld" in litellm.provider_list
+    assert "inworld" in litellm.constants.openai_compatible_providers
+
+
+def test_inworld_provider_resolution(monkeypatch: pytest.MonkeyPatch):
+    from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider
+
+    monkeypatch.setenv("INWORLD_API_KEY", "inworld-test-key")
+
+    model, provider, api_key, api_base = get_llm_provider(
+        model="inworld/inworld/models/gemma-4-31b-it",
+        custom_llm_provider=None,
+        api_base=None,
+        api_key=None,
+    )
+
+    assert model == "inworld/models/gemma-4-31b-it"
+    assert provider == "inworld"
+    assert api_key == "inworld-test-key"
+    assert api_base == "https://api.inworld.ai/v1"
+
+
+def test_inworld_provider_keeps_explicit_credentials(monkeypatch: pytest.MonkeyPatch):
+    from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider
+
+    monkeypatch.setenv("INWORLD_API_KEY", "inworld-env-key")
+    monkeypatch.setenv("INWORLD_API_BASE", "https://inworld.env.example/v1")
+
+    _, provider, api_key, api_base = get_llm_provider(
+        model="inworld/inworld/models/gemma-4-31b-it",
+        custom_llm_provider=None,
+        api_base="https://inworld.internal.example/v1",
+        api_key="inworld-explicit-key",
+    )
+
+    assert provider == "inworld"
+    assert api_key == "inworld-explicit-key"
+    assert api_base == "https://inworld.internal.example/v1"
+
+
+def test_inworld_api_base_env_overrides_default(monkeypatch: pytest.MonkeyPatch):
+    from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider
+
+    monkeypatch.setenv("INWORLD_API_KEY", "inworld-env-key")
+    monkeypatch.setenv("INWORLD_API_BASE", "https://api.eu.inworld.ai/v1")
+
+    _, provider, _, api_base = get_llm_provider(
+        model="inworld/inworld/models/gemma-4-31b-it",
+        custom_llm_provider=None,
+        api_base=None,
+        api_key=None,
+    )
+
+    assert provider == "inworld"
+    assert api_base == "https://api.eu.inworld.ai/v1"
+
+
+def test_inworld_api_base_autodetects_provider(monkeypatch: pytest.MonkeyPatch):
+    from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider
+
+    monkeypatch.setenv("INWORLD_API_KEY", "inworld-env-key")
+
+    model, provider, api_key, api_base = get_llm_provider(
+        model="gemma-4-31b-it",
+        custom_llm_provider=None,
+        api_base="https://api.inworld.ai/v1",
+        api_key=None,
+    )
+
+    assert model == "gemma-4-31b-it"
+    assert provider == "inworld"
+    assert api_key == "inworld-env-key"
+    assert api_base == "https://api.inworld.ai/v1"
+
+
+def test_inworld_is_available_in_add_model_form():
+    fields_path = Path(litellm.__file__).parent / "proxy" / "public_endpoints" / "provider_create_fields.json"
+    providers = json.loads(fields_path.read_text())
+    inworld = next(provider for provider in providers if provider["litellm_provider"] == "inworld")
+
+    assert inworld["provider"] == "INWORLD"
+    assert inworld["provider_display_name"] == "Inworld"
+    assert inworld["default_model_placeholder"] == "inworld/inworld/models/gemma-4-31b-it"
+    assert {field["key"]: field["required"] for field in inworld["credential_fields"]} == {
+        "api_base": False,
+        "api_key": True,
+    }
+
+
+def test_inworld_supported_endpoints():
+    expected: Final = {
+        "chat_completions": True,
+        "messages": True,
+        "responses": False,
+        "embeddings": False,
+        "image_generations": False,
+        "audio_transcriptions": True,
+        "audio_speech": True,
+        "moderations": False,
+        "batches": False,
+        "rerank": False,
+        "a2a": False,
+        "interactions": False,
+    }
+    backup_path = Path(litellm.__file__).parent / "provider_endpoints_support_backup.json"
+    root_path = Path(litellm.__file__).parent.parent / "provider_endpoints_support.json"
+
+    assert json.loads(backup_path.read_text())["providers"]["inworld"]["endpoints"] == expected
+    assert json.loads(root_path.read_text())["providers"]["inworld"]["endpoints"] == expected
+
+
+def test_inworld_chat_completion_request():
+    with respx.mock() as upstream:
+        route: Final = upstream.post("https://api.inworld.ai/v1/chat/completions").respond(200, json=_CHAT_COMPLETION)
+        response: Final = litellm.completion(
+            model="inworld/inworld/models/gemma-4-31b-it",
+            messages=[{"role": "user", "content": "Say hello"}],
+            api_key="inworld-test-key",
+        )
+
+    request: Final = route.calls.last.request
+    body: Final = json.loads(request.content)
+    assert route.call_count == 1
+    assert str(request.url) == "https://api.inworld.ai/v1/chat/completions"
+    assert request.headers["authorization"] == "Bearer inworld-test-key"
+    # Inworld-hosted model IDs live under the "inworld/" namespace upstream; only LiteLLM's routing prefix is removed.
+    assert body["model"] == "inworld/models/gemma-4-31b-it"
+    assert body["messages"] == [{"role": "user", "content": "Say hello"}]
+    assert response.choices[0].message.content == "Hello from Inworld"
+
+
+def test_inworld_routes_third_party_model_ids_unchanged():
+    with respx.mock() as upstream:
+        route: Final = upstream.post("https://api.inworld.ai/v1/chat/completions").respond(200, json=_CHAT_COMPLETION)
+        litellm.completion(
+            model="inworld/openai/gpt-5-mini",
+            messages=[{"role": "user", "content": "Say hello"}],
+            api_key="inworld-test-key",
+        )
+
+    assert json.loads(route.calls.last.request.content)["model"] == "openai/gpt-5-mini"
+
+
+@pytest.mark.asyncio
+async def test_inworld_anthropic_messages_request_is_bridged_to_chat_completions(monkeypatch: pytest.MonkeyPatch):
+    monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)
+    monkeypatch.setattr(litellm, "in_memory_llm_clients_cache", LLMClientCache())
+    with respx.mock() as upstream:
+        route: Final = upstream.post("https://api.inworld.ai/v1/chat/completions").respond(200, json=_CHAT_COMPLETION)
+        response: Final = await litellm.anthropic.messages.acreate(
+            model="inworld/inworld/models/gemma-4-31b-it",
+            messages=[{"role": "user", "content": "Say hello"}],
+            max_tokens=32,
+            api_key="inworld-test-key",
+        )
+
+    request: Final = route.calls.last.request
+    body: Final = json.loads(request.content)
+    assert route.call_count == 1
+    assert request.headers["authorization"] == "Bearer inworld-test-key"
+    assert body["model"] == "inworld/models/gemma-4-31b-it"
+    assert body["max_tokens"] == 32
+    assert response["content"][0]["text"] == "Hello from Inworld"
+
+
+def test_inworld_speech_uses_provider_api_key_from_env(monkeypatch: pytest.MonkeyPatch):
+    monkeypatch.setenv("INWORLD_API_KEY", "inworld-env-key")
+    monkeypatch.delenv("OPENAI_API_KEY", raising=False)
+    monkeypatch.setattr(litellm, "in_memory_llm_clients_cache", LLMClientCache())
+    with respx.mock() as upstream:
+        route: Final = upstream.post("https://api.inworld.ai/v1/audio/speech").respond(
+            200, content=b"ID3-audio", headers={"Content-Type": "audio/mpeg"}
+        )
+        response: Final = litellm.speech(
+            model="inworld/inworld-tts-2",
+            voice="Dennis",
+            input="Hello from Inworld",
+        )
+
+    request: Final = route.calls.last.request
+    body: Final = json.loads(request.content)
+    assert route.call_count == 1
+    assert request.headers["authorization"] == "Bearer inworld-env-key"
+    assert body["model"] == "inworld-tts-2"
+    assert body["voice"] == "Dennis"
+    assert body["input"] == "Hello from Inworld"
+    assert response.content == b"ID3-audio"
+
+
+def test_inworld_transcription_request(monkeypatch: pytest.MonkeyPatch):
+    monkeypatch.setenv("INWORLD_API_KEY", "inworld-env-key")
+    monkeypatch.setattr(litellm, "in_memory_llm_clients_cache", LLMClientCache())
+    with respx.mock() as upstream:
+        route: Final = upstream.post("https://api.inworld.ai/v1/audio/transcriptions").respond(
+            200, json={"text": "Hello from Inworld"}
+        )
+        response: Final = litellm.transcription(
+            model="inworld/inworld/inworld-stt-1",
+            file=("hello.mp3", b"ID3-audio", "audio/mpeg"),
+        )
+
+    request: Final = route.calls.last.request
+    assert route.call_count == 1
+    assert request.headers["authorization"] == "Bearer inworld-env-key"
+    assert b'name="model"\r\n\r\ninworld/inworld-stt-1' in request.content
+    assert response.text == "Hello from Inworld"