Port streaming optimization and X-Cashu refactor from d8a81fb

OpenAI responses api

example.py

@@ -1,37 +0,0 @@
-import os
-
-import openai
-
-client = openai.OpenAI(
-    api_key=os.environ["CASHU_TOKEN"],
-    base_url=os.environ.get("ROUTSTR_API_URL", "https://api.routstr.com/v1"),
-    # base_url="http://roustrjfsdgfiueghsklchg.onion/v1",
-    # client=httpx.AsyncClient(
-    #     proxies={"http": "socks5://localhost:9050"},
-    # ),  # to use onion proxy (tor)
-)
-history: list = []
-
-
-def chat() -> None:
-    while True:
-        user_msg = {"role": "user", "content": input("\nYou: ")}
-        history.append(user_msg)
-        ai_msg = {"role": "assistant", "content": ""}
-
-        for chunk in client.chat.completions.create(
-            model=os.environ.get("MODEL", "openai/gpt-4o-mini"),
-            messages=history,
-            stream=True,
-        ):
-            if len(chunk.choices) > 0:
-                content = chunk.choices[0].delta.content
-                if content is not None:
-                    ai_msg["content"] += content
-                    print(content, end="", flush=True)
-        print()
-        history.append(ai_msg)
-
-
-if __name__ == "__main__":
-    chat()

examples/balance/check_balance.py

@@ -0,0 +1,11 @@
+import os
+
+import httpx
+
+# Use your Cashu token or API key as the Bearer token,
+# cashu token is hashed on the server and acts as an Temporary API key
+headers = {"Authorization": f"Bearer {os.environ.get('TOKEN')}"}
+base_url = os.environ.get("API_URL", "https://api.routstr.com/v1")
+
+resp = httpx.get(f"{base_url}/balance/info", headers=headers)
+print(resp.json())

examples/balance/create_balance.py

@@ -0,0 +1,15 @@
+import os
+
+import httpx
+
+# Send a Cashu token to the /create endpoint to get a persistent API key
+token = os.environ.get("TOKEN")
+if not token:
+    print("Please set TOKEN environment variable with a Cashu token")
+    exit(1)
+
+base_url = os.environ.get("API_URL", "https://api.routstr.com/v1")
+
+resp = httpx.get(f"{base_url}/balance/create", params={"initial_balance_token": token})
+
+print(resp.json())

examples/balance/refund_balance.py

@@ -0,0 +1,12 @@
+import os
+
+import httpx
+
+# Use your Cashu token or API key as the Bearer token
+headers = {"Authorization": f"Bearer {os.environ.get('TOKEN')}"}
+base_url = os.environ.get("API_URL", "https://api.routstr.com/v1")
+
+resp = httpx.post(f"{base_url}/balance/refund", headers=headers)
+
+print("Refund successful!")
+print(resp.json())

examples/balance/topup_balance.py

@@ -0,0 +1,16 @@
+import os
+
+import httpx
+
+# Use your Cashu token or API key as the Bearer token
+headers = {"Authorization": f"Bearer {os.environ.get('TOKEN')}"}
+base_url = os.environ.get("API_URL", "https://api.routstr.com/v1")
+
+# The Cashu token to top up with
+cashu_token = input("Enter Cashu token to top up: ")
+
+resp = httpx.post(
+    f"{base_url}/balance/topup", headers=headers, json={"cashu_token": cashu_token}
+)
+
+print(resp.json())

examples/chat_completions.py

@@ -0,0 +1,15 @@
+import os
+
+from openai import OpenAI
+
+client = OpenAI(
+    api_key=os.environ.get("TOKEN"),
+    base_url=os.environ.get("API_URL", "https://api.routstr.com/v1"),
+)
+
+response = client.chat.completions.create(
+    model=os.environ.get("MODEL", "gpt-5-nano"),
+    messages=[{"role": "user", "content": "Hello!"}],
+)
+
+print(response.choices[0].message.content)

examples/list_models.py

@@ -0,0 +1,19 @@
+import os
+
+import httpx
+from openai import OpenAI
+
+client = OpenAI(
+    api_key=os.environ.get("TOKEN", ""),
+    base_url=os.environ.get("API_URL", "https://api.routstr.com/v1"),
+)
+
+for model in client.models.list():
+    print(model.id)
+
+# OR
+
+models = httpx.get(
+    f"{client.base_url}/v1/models",
+    headers={"Authorization": f"Bearer {client.api_key}"},
+).json()

examples/responses/conversation.py

@@ -0,0 +1,31 @@
+import os
+
+from openai import OpenAI
+
+client = OpenAI(
+    api_key=os.environ.get("TOKEN"),
+    base_url=os.environ.get("API_URL", "https://api.routstr.com/v1"),
+)
+
+conversation = []  # type: ignore
+
+# First turn
+response1 = client.responses.create(  # type: ignore
+    model="o4-mini",
+    input="Hi, my name is Alice.",
+    conversation=conversation,
+)
+print("Response 1:", response1.output)
+
+# Note: The 'conversation' parameter might need to be constructed differently
+# depending on exact SDK/API spec. Typically, you pass back the previous turn's data.
+# Assuming the SDK manages or returns a conversation object/ID:
+# conversation.append(response1)
+
+# Second turn - demonstrating intent, actual implementation depends on strict API spec
+# response2 = client.responses.create(
+#     model="openai/gpt-4o-mini",
+#     input="What is my name?",
+#     conversation=conversation,
+# )
+# print("Response 2:", response2.output)

examples/responses/create.py

@@ -0,0 +1,17 @@
+import os
+
+from openai import OpenAI
+
+# The OpenAI SDK handles the 'responses' endpoint if it's updated to the latest version
+# and the base_url points to a compatible proxy like Routstr.
+client = OpenAI(
+    api_key=os.environ.get("TOKEN"),
+    base_url=os.environ.get("API_URL", "https://api.routstr.com/v1"),
+)
+
+response = client.responses.create(
+    model="gpt-5-mini",
+    input="Tell me a three sentence bedtime story about a unicorn.",
+)
+
+print(response.output)

examples/responses/streaming_response.py

@@ -0,0 +1,20 @@
+import os
+
+from openai import OpenAI
+
+client = OpenAI(
+    api_key=os.environ.get("TOKEN"),
+    base_url=os.environ.get("API_URL", "https://api.routstr.com/v1"),
+)
+
+stream = client.responses.create(
+    model="claude-4.5-sonnet",
+    input="Write a short poem about rust.",
+    stream=True,
+)
+
+for event in stream:
+    # Note: Depending on the SDK version and response structure,
+    # you might access event.output_delta or similar fields
+    print(event, end="", flush=True)
+print()

examples/responses/web_search.py

@@ -0,0 +1,16 @@
+import os
+
+from openai import OpenAI
+
+client = OpenAI(
+    api_key=os.environ.get("TOKEN"),
+    base_url=os.environ.get("API_URL", "https://api.routstr.com/v1"),
+)
+
+response = client.responses.create(
+    model="gpt-5-mini",
+    input="What is the latest news about AI?",
+    tools=[{"type": "web_search"}],  # type: ignore
+)
+
+print(response.output)

examples/streaming.py

@@ -0,0 +1,28 @@
+import os
+
+from openai import OpenAI
+
+client = OpenAI(
+    api_key=os.environ.get("TOKEN"),
+    base_url=os.environ.get("API_URL", "https://api.routstr.com/v1"),
+)
+
+messages = []
+while True:
+    messages.append({"role": "user", "content": input("\nYou: ")})
+
+    stream = client.chat.completions.create(
+        model=os.environ.get("MODEL", "gpt-5.1-mini"),
+        messages=messages,  # type: ignore
+        stream=True,
+    )
+
+    print("AI: ", end="")
+    response_content = ""
+    for chunk in stream:
+        if content := chunk.choices[0].delta.content:  # type: ignore
+            print(content, end="", flush=True)
+            response_content += content
+    print()
+
+    messages.append({"role": "assistant", "content": response_content})

examples/tor.py

@@ -0,0 +1,20 @@
+import os
+
+import httpx
+from openai import OpenAI
+
+# Requires `pip install "httpx[socks]"` and a running Tor proxy on port 9050
+client = OpenAI(
+    api_key=os.environ.get("TOKEN"),
+    base_url=os.environ.get("ONION_URL", "http://roustrjfsdgfiueghsklchg.onion/v1"),
+    http_client=httpx.Client(proxies="socks5://localhost:9050"),
+)
+
+print(
+    client.chat.completions.create(
+        model="openai/gpt-4o-mini",
+        messages=[{"role": "user", "content": "Hello from Tor!"}],
+    )
+    .choices[0]
+    .message.content
+)

pyproject.toml

@@ -73,6 +73,7 @@ packages = ["routstr"]
 [tool.ruff.lint]
 select = ["E", "F", "I"]
 ignore = ["E501"]
+exclude = ["examples"]
 
 [tool.mypy]
 python_version = "3.11"

routstr/core/settings.py

@@ -6,7 +6,7 @@ import os
 from datetime import datetime, timezone
 from typing import Any
 
-from pydantic.v1 import BaseModel, BaseSettings, Field
+from pydantic.v1 import BaseModel, BaseSettings, Field, validator
 from sqlmodel.ext.asyncio.session import AsyncSession
 
 
@@ -37,6 +37,13 @@ class Settings(BaseSettings):
 
     # Cashu
     cashu_mints: list[str] = Field(default_factory=list, env="CASHU_MINTS")
+
+    @validator("cashu_mints", pre=True, each_item=True)
+    def normalize_mint_url(cls, v: str) -> str:
+        if isinstance(v, str):
+            return v.rstrip("/")
+        return v
+
     receive_ln_address: str = Field(default="", env="RECEIVE_LN_ADDRESS")
     primary_mint: str = Field(default="", env="PRIMARY_MINT_URL")
     primary_mint_unit: str = Field(default="sat", env="PRIMARY_MINT_UNIT")

routstr/payment/cost_calculation.py

@@ -191,6 +191,10 @@ async def calculate_cost(  # todo: can be sync
         output_tokens if output_tokens != 0 else usage_data.get("output_tokens", 0)
     )
 
+    # added for response api
+    input_tokens = input_tokens if input_tokens != 0 else  response_data.get("usage", {}).get("input_tokens", 0)
+    output_tokens = output_tokens if output_tokens != 0 else  response_data.get("usage", {}).get("output_tokens", 0)
+
     input_msats = round(input_tokens / 1000 * MSATS_PER_1K_INPUT_TOKENS, 3)
 
     output_msats = round(output_tokens / 1000 * MSATS_PER_1K_OUTPUT_TOKENS, 3)

routstr/proxy.py

@@ -16,6 +16,7 @@ from .core.db import (
     create_session,
     get_session,
 )
+from .core.settings import settings
 from .payment.helpers import (
     calculate_discounted_max_cost,
     check_token_balance,
@@ -25,6 +26,7 @@ from .payment.helpers import (
 from .payment.models import Model
 from .upstream import BaseUpstreamProvider
 from .upstream.helpers import init_upstreams
+from .wallet import deserialize_token_from_string
 
 logger = get_logger(__name__)
 proxy_router = APIRouter()
@@ -137,20 +139,32 @@ async def proxy(
             "unauthorized", "Unauthorized", 401, request=request
         )
 
-    logger.info(  # TODO: move to middleware, async
-        "Received proxy request",
-        extra={
-            "method": request.method,
-            "path": path,
-            "client_host": request.client.host if request.client else "unknown",
-            "user_agent": request.headers.get("user-agent", "unknown")[:100],
-        },
-    )
-
+    is_responses_api = path.startswith("v1/responses") or path.startswith("responses")
     request_body = await request.body()
     request_body_dict = parse_request_body_json(request_body, path)
 
-    model_id = request_body_dict.get("model", "unknown")
+    if is_responses_api:
+        model_id = extract_model_from_responses_request(request_body_dict)
+    else:
+        model_id = request_body_dict.get("model", "unknown")
+
+    if "https://testnut.cashu.space" in settings.cashu_mints:
+        try:
+            token_str = None
+            if x_cashu_header := headers.get("x-cashu"):
+                token_str = x_cashu_header
+            elif auth_header := headers.get("authorization"):
+                parts = auth_header.split(" ")
+                if len(parts) > 1 and not parts[1].startswith("sk-"):
+                    token_str = parts[1]
+
+            if token_str:
+                token_obj = deserialize_token_from_string(token_str)
+                if token_obj.mint == "https://testnut.cashu.space":
+                    model_id = "mock/gpt-420-mock"
+                    request_body_dict["model"] = model_id
+        except Exception:
+            pass
 
     model_obj = get_model_instance(model_id)
     if not model_obj:
@@ -176,9 +190,14 @@ async def proxy(
     check_token_balance(headers, request_body_dict, max_cost_for_model)
 
     if x_cashu := headers.get("x-cashu", None):
-        return await upstream.handle_x_cashu(
-            request, x_cashu, path, max_cost_for_model, model_obj
-        )
+        if is_responses_api:
+            return await upstream.handle_x_cashu_responses(
+                request, x_cashu, path, max_cost_for_model, model_obj
+            )
+        else:
+            return await upstream.handle_x_cashu(
+                request, x_cashu, path, max_cost_for_model, model_obj
+            )
 
     elif auth := headers.get("authorization", None):
         key = await get_bearer_token_key(headers, path, session, auth)
@@ -193,28 +212,36 @@ async def proxy(
             )
 
         logger.debug("Processing unauthenticated GET request", extra={"path": path})
-        # TODO: why is this needed? can we remove it?
         headers = upstream.prepare_headers(dict(request.headers))
         return await upstream.forward_get_request(request, path, headers)
 
-    # Only pay for request if we have request body data (for completions endpoints)
     if request_body_dict:
         await pay_for_request(key, max_cost_for_model, session)
 
-    # Prepare headers for upstream
     headers = upstream.prepare_headers(dict(request.headers))
 
-    # Forward to upstream and handle response
-    response = await upstream.forward_request(
-        request,
-        path,
-        headers,
-        request_body,
-        key,
-        max_cost_for_model,
-        session,
-        model_obj,
-    )
+    if is_responses_api:
+        response = await upstream.forward_responses_request(
+            request,
+            path,
+            headers,
+            request_body,
+            key,
+            max_cost_for_model,
+            session,
+            model_obj,
+        )
+    else:
+        response = await upstream.forward_request(
+            request,
+            path,
+            headers,
+            request_body,
+            key,
+            max_cost_for_model,
+            session,
+            model_obj,
+        )
 
     if response.status_code != 200:
         await revert_pay_for_request(key, session, max_cost_for_model)
@@ -317,6 +344,24 @@ async def get_bearer_token_key(
         raise
 
 
+def extract_model_from_responses_request(request_body_dict: dict[str, Any]) -> str:
+    if model := request_body_dict.get("model"):
+        return model
+
+    if input_data := request_body_dict.get("input"):
+        if isinstance(input_data, dict) and (model := input_data.get("model")):
+            return model
+
+    if request_body_dict.get("messages"):
+        return "unknown"
+
+    logger.warning(
+        "No model found in Responses API request",
+        extra={"body_keys": list(request_body_dict.keys())}
+    )
+    return "unknown"
+
+
 def parse_request_body_json(request_body: bytes, path: str) -> dict[str, Any]:
     request_body_dict = {}
     if request_body:

routstr/upstream/fake.py

@@ -0,0 +1,265 @@
+import asyncio
+import json
+import random
+from typing import AsyncIterator
+
+from fastapi import Request
+from fastapi.responses import Response, StreamingResponse
+
+from ..core.db import ApiKey, AsyncSession
+from ..payment.models import Architecture, Model, Pricing
+from .base import BaseUpstreamProvider
+
+
+class MockUpstreamProvider(BaseUpstreamProvider):
+    """Fack Mock Upstream provider specifically for Testing."""
+
+    provider_type = "mock"
+
+    async def forward_request(
+        self,
+        request: Request,
+        path: str,
+        headers: dict,
+        request_body: bytes | None,
+        key: ApiKey,
+        max_cost_for_model: int,
+        session: AsyncSession,
+        model_obj: Model,
+    ) -> Response | StreamingResponse:
+        if path.endswith("chat/completions"):
+            is_streaming = False
+            if request_body:
+                request_data = json.loads(request_body)
+                is_streaming = request_data.get("stream", False)
+
+            if is_streaming:
+
+                async def fake_streaming_response(
+                    chunk_size: int | None = None,
+                ) -> AsyncIterator[bytes]:
+                    suffix = random.randint(1000, 9999)
+                    req_id = f"gen-mock-stream-{suffix}"
+                    created = 1766138895
+                    model = "mock/gpt-420-mock"
+
+                    def make_chunk(
+                        delta: dict,
+                        finish_reason: str | None = None,
+                        usage: dict | None = None,
+                    ) -> bytes:
+                        chunk = {
+                            "id": req_id,
+                            "provider": "MockProvider",
+                            "model": model,
+                            "object": "chat.completion.chunk",
+                            "created": created,
+                            "choices": [
+                                {
+                                    "index": 0,
+                                    "delta": delta,
+                                    "finish_reason": finish_reason,
+                                    "native_finish_reason": "completed"
+                                    if finish_reason
+                                    else None,
+                                    "logprobs": None,
+                                }
+                            ],
+                        }
+                        if usage:
+                            chunk["usage"] = usage
+                        return f"data: {json.dumps(chunk)}\n\n".encode()
+
+                    # 1. Initial chunk
+                    yield make_chunk({"role": "assistant", "content": ""})
+                    await asyncio.sleep(0.02)
+
+                    # 2. Reasoning chunks
+                    reasoning_tokens = ["Mock", " reason", "ing", "..."]
+                    for token in reasoning_tokens:
+                        delta = {
+                            "role": "assistant",
+                            "content": "",
+                            "reasoning": token,
+                            "reasoning_details": [
+                                {
+                                    "type": "reasoning.summary",
+                                    "summary": token,
+                                    "format": "openai-responses-v1",
+                                    "index": 0,
+                                }
+                            ],
+                        }
+                        yield make_chunk(delta)
+                        await asyncio.sleep(0.03)
+
+                    # 3. Content chunks
+                    content_tokens = ["This", " is", " a", " mock", " stream", "."]
+                    for token in content_tokens:
+                        yield make_chunk({"role": "assistant", "content": token})
+                        await asyncio.sleep(0.03)
+
+                    # 4. Finish chunk
+                    yield make_chunk(
+                        {"role": "assistant", "content": ""}, finish_reason="stop"
+                    )
+
+                    # 5. Usage chunk
+                    usage_data = {
+                        "prompt_tokens": 10,
+                        "completion_tokens": 20,
+                        "total_tokens": 30,
+                        "cost": 0.001,
+                        "is_byok": False,
+                        "prompt_tokens_details": {
+                            "cached_tokens": 0,
+                            "audio_tokens": 0,
+                            "video_tokens": 0,
+                        },
+                        "cost_details": {
+                            "upstream_inference_cost": None,
+                            "upstream_inference_prompt_cost": 0,
+                            "upstream_inference_completions_cost": 0.001,
+                        },
+                        "completion_tokens_details": {
+                            "reasoning_tokens": 10,
+                            "image_tokens": 0,
+                        },
+                    }
+
+                    usage_chunk = {
+                        "id": req_id,
+                        "provider": "MockProvider",
+                        "model": model,
+                        "object": "chat.completion.chunk",
+                        "created": created,
+                        "choices": [
+                            {
+                                "index": 0,
+                                "delta": {"role": "assistant", "content": ""},
+                                "finish_reason": None,
+                                "native_finish_reason": None,
+                                "logprobs": None,
+                            }
+                        ],
+                        "usage": usage_data,
+                    }
+                    yield f"data: {json.dumps(usage_chunk)}\n\n".encode()
+
+                    # 6. DONE
+                    yield b"data: [DONE]\n\n"
+
+                    # 7. Cost
+                    cost_chunk = {
+                        "cost": {
+                            "base_msats": 0,
+                            "input_msats": 2,
+                            "output_msats": 10,
+                            "total_msats": 12,
+                        }
+                    }
+                    yield f"data: {json.dumps(cost_chunk)}\n\n".encode()
+
+                return StreamingResponse(
+                    fake_streaming_response(),
+                    200,
+                )
+
+            else:
+                suffix = random.randint(1000, 9999)
+                content_dict = {
+                    "id": f"gen-mock-{suffix}",
+                    "provider": "MockProvider",
+                    "model": "mock/gpt-5-mini",
+                    "object": "chat.completion",
+                    "created": 1766138655,
+                    "choices": [
+                        {
+                            "logprobs": None,
+                            "finish_reason": "length",
+                            "native_finish_reason": "max_output_tokens",
+                            "index": 0,
+                            "message": {
+                                "role": "assistant",
+                                "content": f"Mock Content {suffix}",
+                                "refusal": None,
+                                "reasoning": f"Mock Reasoning {suffix}",
+                                "reasoning_details": [
+                                    {
+                                        "format": "openai-responses-v1",
+                                        "index": 0,
+                                        "type": "reasoning.summary",
+                                        "summary": f"Mock Summary {suffix}",
+                                    },
+                                    {
+                                        "id": f"rs_mock_{suffix}",
+                                        "format": "openai-responses-v1",
+                                        "index": 0,
+                                        "type": "reasoning.encrypted",
+                                        "data": "mock_encrypted_data",
+                                    },
+                                ],
+                            },
+                        }
+                    ],
+                    "usage": {
+                        "prompt_tokens": 10,
+                        "completion_tokens": 10,
+                        "total_tokens": 20,
+                        "cost": 0,
+                        "is_byok": False,
+                        "prompt_tokens_details": {
+                            "cached_tokens": 0,
+                            "audio_tokens": 0,
+                            "video_tokens": 0,
+                        },
+                        "cost_details": {
+                            "upstream_inference_cost": None,
+                            "upstream_inference_prompt_cost": 0,
+                            "upstream_inference_completions_cost": 0,
+                        },
+                        "completion_tokens_details": {
+                            "reasoning_tokens": 5,
+                            "image_tokens": 0,
+                        },
+                    },
+                    "cost": {
+                        "base_msats": 0,
+                        "input_msats": 0,
+                        "output_msats": 0,
+                        "total_msats": 0,
+                    },
+                }
+                return Response(json.dumps(content_dict).encode(), 200)
+
+        elif path.endswith("embeddings"):
+            raise NotImplementedError
+        elif path.endswith("responses"):
+            raise NotImplementedError
+        else:
+            raise NotImplementedError
+
+    async def fetch_models(self) -> list[Model]:
+        return [
+            Model(
+                id="mock/gpt-420-mock",
+                name="mock/gpt-420-mock",
+                created=0,
+                description="mock model for testing",
+                context_length=8192,
+                architecture=Architecture(
+                    modality="text",
+                    input_modalities=["text"],
+                    output_modalities=["text"],
+                    tokenizer="",
+                    instruct_type=None,
+                ),
+                pricing=Pricing(prompt=0.01, completion=0.01),
+            ),
+        ]
+
+    def transform_model_name(self, model_id: str) -> str:
+        return "fake-model"
+
+    async def get_balance(self) -> float | None:
+        return 420.69

routstr/upstream/helpers.py

@@ -218,6 +218,14 @@ async def init_upstreams() -> list[BaseUpstreamProvider]:
         results = await asyncio.gather(*tasks)
         upstreams = [p for p in results if p is not None]
 
+        if "https://testnut.cashu.space" in settings.cashu_mints:
+            from .fake import MockUpstreamProvider
+
+            mock_provider = MockUpstreamProvider("mock", "mock")
+            await mock_provider.refresh_models_cache()
+            upstreams.append(mock_provider)
+            logger.info("Initialized MockUpstreamProvider for testnut mint")
+
         return upstreams
 
 

routstr/wallet.py

@@ -313,6 +313,8 @@ async def periodic_payout() -> None:
         try:
             async with db.create_session() as session:
                 for mint_url in settings.cashu_mints:
+                    if mint_url == "https://testnut.cashu.space":
+                        continue
                     for unit in ["sat", "msat"]:
                         wallet = await get_wallet(mint_url, unit)
                         proofs = get_proofs_per_mint_and_unit(