argo-proxy 3.2.1__tar.gz → 3.2.3__tar.gz

{argo_proxy-3.2.1/src/argo_proxy.egg-info → argo_proxy-3.2.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: argo-proxy
-Version: 3.2.1
+Version: 3.2.3
 Summary: Proxy server to Argo API, OpenAI format compatible
 Author-email: Peng Ding <oaklight@gmx.com>
 License-Expression: MIT
@@ -15,7 +15,7 @@ Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: aiohttp>=3.12.2
-Requires-Dist: llm-rosetta>=0.7.0
+Requires-Dist: llm-rosetta>=0.7.1
 Requires-Dist: pydantic>=2.11.7
 Requires-Dist: tiktoken>=0.9.0
 Requires-Dist: tqdm>=4.67.1

{argo_proxy-3.2.1 → argo_proxy-3.2.3}/pyproject.toml

@@ -20,7 +20,7 @@ classifiers = [
 
 dependencies = [
     "aiohttp>=3.12.2",
-    "llm-rosetta>=0.7.0",
+    "llm-rosetta>=0.7.1",
     "pydantic>=2.11.7",
     "tiktoken>=0.9.0",
     "tqdm>=4.67.1",

{argo_proxy-3.2.1 → argo_proxy-3.2.3/src/argo_proxy.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: argo-proxy
-Version: 3.2.1
+Version: 3.2.3
 Summary: Proxy server to Argo API, OpenAI format compatible
 Author-email: Peng Ding <oaklight@gmx.com>
 License-Expression: MIT
@@ -15,7 +15,7 @@ Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: aiohttp>=3.12.2
-Requires-Dist: llm-rosetta>=0.7.0
+Requires-Dist: llm-rosetta>=0.7.1
 Requires-Dist: pydantic>=2.11.7
 Requires-Dist: tiktoken>=0.9.0
 Requires-Dist: tqdm>=4.67.1

{argo_proxy-3.2.1 → argo_proxy-3.2.3}/src/argo_proxy.egg-info/SOURCES.txt

@@ -9,7 +9,6 @@ src/argo_proxy.egg-info/requires.txt
 src/argo_proxy.egg-info/top_level.txt
 src/argoproxy/__init__.py
 src/argoproxy/app.py
-src/argoproxy/models.py
 src/argoproxy/performance.py
 src/argoproxy/py.typed
 src/argoproxy/_vendor/__init__.py
@@ -28,11 +27,14 @@ src/argoproxy/endpoints/dev_proxy.py
 src/argoproxy/endpoints/dispatch.py
 src/argoproxy/endpoints/extras.py
 src/argoproxy/endpoints/passthrough.py
+src/argoproxy/models/__init__.py
+src/argoproxy/models/constants.py
+src/argoproxy/models/registry.py
+src/argoproxy/models/upstream.py
 src/argoproxy/utils/attack_logger.py
 src/argoproxy/utils/image_processing.py
 src/argoproxy/utils/logging.py
 src/argoproxy/utils/misc.py
-src/argoproxy/utils/models.py
 src/argoproxy/utils/transports.py
 test/test_chat_completions.py
 test/test_embeddings.py
@@ -41,4 +43,5 @@ test/test_function_calling_single.py
 test/test_leaked_tool_parser.py
 test/test_model_resolution.py
 tests/test_config_migrate.py
+tests/test_security_middleware.py
 tests/test_unix_socket.py

{argo_proxy-3.2.1 → argo_proxy-3.2.3}/src/argo_proxy.egg-info/requires.txt

@@ -1,5 +1,5 @@
 aiohttp>=3.12.2
-llm-rosetta>=0.7.0
+llm-rosetta>=0.7.1
 pydantic>=2.11.7
 tiktoken>=0.9.0
 tqdm>=4.67.1

argo_proxy-3.2.3/src/argoproxy/__init__.py

@@ -0,0 +1 @@
+__version__ = "3.2.3"

{argo_proxy-3.2.1 → argo_proxy-3.2.3}/src/argoproxy/app.py

@@ -19,6 +19,7 @@ from .performance import (
     OptimizedHTTPSession,
     get_performance_config,
 )
+from .utils.attack_logger import security_middleware
 from .utils.logging import log_debug, log_error, log_info, log_warning
 
 
@@ -320,7 +321,10 @@ def create_app():
     # Set client_max_size to 100MB to handle large image payloads from remote clients
     # Users may send images larger than the gateway's 20MB limit; argo-proxy will
     # compress them before forwarding. Default aiohttp limit is 1MB which is too small.
-    app = web.Application(client_max_size=100 * 1024 * 1024)
+    app = web.Application(
+        client_max_size=100 * 1024 * 1024,
+        middlewares=[security_middleware],
+    )
     app.on_startup.append(prepare_app)
     app.on_shutdown.append(cleanup_app)
 

{argo_proxy-3.2.1 → argo_proxy-3.2.3}/src/argoproxy/cli/handlers.py

@@ -700,7 +700,7 @@ def handle_models(args: argparse.Namespace):
     from collections import defaultdict
 
     from ..config import load_config
-    from ..models import ModelRegistry
+    from ..models import ModelRegistry, classify_model_family
 
     config_data, _ = load_config(args.config, verbose=False)
     if not config_data:
@@ -762,7 +762,7 @@ def handle_models(args: argparse.Namespace):
     if args.json:
         output = []
         for internal_id, aliases in sorted(chat_id_to_aliases.items()):
-            family = registry._classify_model_by_family(internal_id)
+            family = classify_model_family(internal_id)
             output.append(
                 {
                     "upstream_id": internal_id,
@@ -807,7 +807,7 @@ def handle_models(args: argparse.Namespace):
     # Classify chat models by family
     chat_families: dict[str, list[tuple[str, list[str]]]] = defaultdict(list)
     for internal_id, aliases in sorted(chat_id_to_aliases.items()):
-        family = registry._classify_model_by_family(internal_id)
+        family = classify_model_family(internal_id)
         chat_families[family].append((internal_id, aliases))
 
     for family in family_order:

{argo_proxy-3.2.1 → argo_proxy-3.2.3}/src/argoproxy/endpoints/dispatch.py

@@ -27,6 +27,7 @@ from llm_rosetta.shims.providers import load_providers
 from ..config import ArgoConfig
 from ..models import ModelRegistry
 from ..utils.image_processing import process_anthropic_images, process_openai_images
+from ..utils.misc import build_user_agent
 from ..utils.logging import (
     clear_request_user,
     log_converted_request,
@@ -288,10 +289,7 @@ def _build_upstream_headers(
     """Build HTTP headers for the upstream API request."""
     headers: dict[str, str] = {"Content-Type": "application/json"}
 
-    # Pass through client's User-Agent to upstream (transparent proxy)
-    client_ua = request.headers.get("User-Agent")
-    if client_ua:
-        headers["User-Agent"] = client_ua
+    headers["User-Agent"] = build_user_agent(request.headers.get("User-Agent"))
 
     if should_use_username_passthrough():
         api_key = _extract_client_credential(request, target_provider) or fallback_user

{argo_proxy-3.2.1 → argo_proxy-3.2.3}/src/argoproxy/endpoints/passthrough.py

@@ -13,6 +13,7 @@ from aiohttp import web
 
 from ..config import ArgoConfig
 from ..models import ModelRegistry
+from ..utils.misc import build_user_agent
 from ..utils.logging import (
     log_converted_request,
     log_debug,
@@ -67,6 +68,9 @@ async def proxy_embeddings_request(
         upstream_url = f"{config.native_openai_base_url}/embeddings"
 
         headers = {"Content-Type": "application/json"}
+
+        headers["User-Agent"] = build_user_agent(request.headers.get("User-Agent"))
+
         if "Authorization" in request.headers:
             headers["Authorization"] = request.headers["Authorization"]
 

argo_proxy-3.2.3/src/argoproxy/models/__init__.py

@@ -0,0 +1,45 @@
+"""Model registry, classification, and upstream communication."""
+
+from .constants import (
+    ANTHROPIC_PATTERNS,
+    GEMINI_PATTERN,
+    GPT_O_PATTERN,
+    NATIVE_TOOL_CALL_MODELS,
+    NATIVE_TOOL_CALL_PATTERNS,
+    NO_SYS_MSG_MODELS,
+    NO_SYS_MSG_PATTERNS,
+    OPTION_2_INPUT_MODELS,
+    OPTION_2_INPUT_PATTERNS,
+    TIKTOKEN_ENCODING_PREFIX_MAPPING,
+    _DEFAULT_CHAT_MODELS,
+    _EMBED_MODELS,
+    classify_model_family,
+    filter_model_by_patterns,
+    flatten_mapping,
+    is_anthropic_model,
+)
+from .registry import ModelRegistry, OpenAIModel
+from .upstream import Model, produce_argo_model_list
+
+__all__ = [
+    "ANTHROPIC_PATTERNS",
+    "GEMINI_PATTERN",
+    "GPT_O_PATTERN",
+    "NATIVE_TOOL_CALL_MODELS",
+    "NATIVE_TOOL_CALL_PATTERNS",
+    "NO_SYS_MSG_MODELS",
+    "NO_SYS_MSG_PATTERNS",
+    "OPTION_2_INPUT_MODELS",
+    "OPTION_2_INPUT_PATTERNS",
+    "TIKTOKEN_ENCODING_PREFIX_MAPPING",
+    "Model",
+    "ModelRegistry",
+    "OpenAIModel",
+    "_DEFAULT_CHAT_MODELS",
+    "_EMBED_MODELS",
+    "classify_model_family",
+    "filter_model_by_patterns",
+    "flatten_mapping",
+    "is_anthropic_model",
+    "produce_argo_model_list",
+]

argo_proxy-3.2.3/src/argoproxy/models/constants.py

@@ -0,0 +1,153 @@
+"""Model constants, default lists, patterns, and classification helpers."""
+
+import fnmatch
+from typing import Any
+
+
+def flatten_mapping(mapping: dict[str, Any]) -> dict[str, str]:
+    flat = {}
+    for model, aliases in mapping.items():
+        if isinstance(aliases, str):
+            flat[aliases] = model
+        else:
+            for alias in aliases:
+                flat[alias] = model
+    return flat
+
+
+# Default models fallback — kept in sync with the live /v1/models endpoint.
+_DEFAULT_CHAT_MODELS = flatten_mapping(
+    {
+        # openai – gpt-4o (legacy, still served)
+        "gpt4o": "argo:gpt-4o",
+        # openai – o-series reasoning
+        "gpto1": ["argo:gpt-o1", "argo:o1"],
+        "gpto3mini": ["argo:gpt-o3-mini", "argo:o3-mini"],
+        "gpto3": ["argo:gpt-o3", "argo:o3"],
+        "gpto4mini": ["argo:gpt-o4-mini", "argo:o4-mini"],
+        # openai – gpt-4.1 family
+        "gpt41": "argo:gpt-4.1",
+        "gpt41mini": "argo:gpt-4.1-mini",
+        "gpt41nano": "argo:gpt-4.1-nano",
+        # openai – gpt-5 family
+        "gpt5": "argo:gpt-5",
+        "gpt5mini": "argo:gpt-5-mini",
+        "gpt5nano": "argo:gpt-5-nano",
+        "gpt51": "argo:gpt-5.1",
+        "gpt52": "argo:gpt-5.2",
+        "gpt54": "argo:gpt-5.4",
+        "gpt54mini": "argo:gpt-5.4-mini",
+        "gpt54nano": "argo:gpt-5.4-nano",
+        "gpt55": "argo:gpt-5.5",
+        # gemini
+        "gemini25pro": "argo:gemini-2.5-pro",
+        "gemini25flash": "argo:gemini-2.5-flash",
+        "gemini35flash": "argo:gemini-3.5-flash",
+        "gemini31flashlite": "argo:gemini-3.1-flash-lite",
+        # claude – opus
+        "claudeopus48": ["argo:claude-4.8-opus", "argo:claude-opus-4.8"],
+        "claudeopus47": ["argo:claude-4.7-opus", "argo:claude-opus-4.7"],
+        "claudeopus46": ["argo:claude-4.6-opus", "argo:claude-opus-4.6"],
+        "claudeopus45": ["argo:claude-4.5-opus", "argo:claude-opus-4.5"],
+        "claudeopus41": ["argo:claude-4.1-opus", "argo:claude-opus-4.1"],
+        # claude – sonnet
+        "claudesonnet5": ["argo:claude-5-sonnet", "argo:claude-sonnet-5"],
+        "claudesonnet46": ["argo:claude-4.6-sonnet", "argo:claude-sonnet-4.6"],
+        "claudesonnet45": ["argo:claude-4.5-sonnet", "argo:claude-sonnet-4.5"],
+        # claude – haiku
+        "claudehaiku45": ["argo:claude-4.5-haiku", "argo:claude-haiku-4.5"],
+    }
+)
+
+_EMBED_MODELS = flatten_mapping(
+    {
+        "ada002": "argo:text-embedding-ada-002",
+        "v3small": "argo:text-embedding-3-small",
+        "v3large": "argo:text-embedding-3-large",
+    }
+)
+
+
+def filter_model_by_patterns(
+    model_dict: dict[str, str], patterns: set[str]
+) -> list[str]:
+    """Filter model_dict values (model_id) by given fnmatch patterns,
+    returning both the model_name (key) and model_id (value) for matches."""
+    matching = set()
+    for model_name, model_id in model_dict.items():
+        if any(fnmatch.fnmatch(model_id, pattern) for pattern in patterns):
+            matching.add(model_name)
+            matching.add(model_id)
+    return sorted(matching)
+
+
+# any models that unable to handle system prompt (o1-mini / o1-preview retired)
+NO_SYS_MSG_PATTERNS: set[str] = set()
+
+NO_SYS_MSG_MODELS = filter_model_by_patterns(
+    _DEFAULT_CHAT_MODELS,
+    NO_SYS_MSG_PATTERNS,
+)
+
+
+# any models that only able to handle single system prompt and no system prompt at all
+OPTION_2_INPUT_PATTERNS: set[str] = set()
+
+OPTION_2_INPUT_MODELS = filter_model_by_patterns(
+    _DEFAULT_CHAT_MODELS,
+    OPTION_2_INPUT_PATTERNS,
+)
+
+# any models that supports native tool call
+NATIVE_TOOL_CALL_PATTERNS: set[str] = {
+    "*o1",
+    "*o3*",
+    "*o4*",
+}
+
+NATIVE_TOOL_CALL_MODELS = filter_model_by_patterns(
+    _DEFAULT_CHAT_MODELS,
+    NATIVE_TOOL_CALL_PATTERNS,
+)
+
+TIKTOKEN_ENCODING_PREFIX_MAPPING = {
+    "gpt5": "o200k_base",  # gpt-5 family
+    "gpto": "o200k_base",  # o-series
+    "gpt4o": "o200k_base",  # gpt-4o
+    "gpt41": "o200k_base",  # gpt-4.1 family
+    # this order need to be preserved to correctly parse mapping
+    "gpt4": "cl100k_base",  # gpt-4 series (legacy fallback)
+    "ada002": "cl100k_base",  # embedding
+    "v3": "cl100k_base",  # embedding
+}
+
+# --- Provider detection patterns ---
+
+GPT_O_PATTERN = "gpto*"
+GEMINI_PATTERN = "gemini*"
+
+ANTHROPIC_PATTERNS = ("claude*", "sonnet*", "opus*", "haiku*", "fable*")
+
+
+def is_anthropic_model(model_id: str) -> bool:
+    return any(fnmatch.fnmatch(model_id, p) for p in ANTHROPIC_PATTERNS)
+
+
+def classify_model_family(model_id: str) -> str:
+    """Classify a model by its provider family based on model ID patterns."""
+    if (
+        fnmatch.fnmatch(model_id, "gpt*")
+        or fnmatch.fnmatch(model_id, GPT_O_PATTERN)
+        or fnmatch.fnmatch(model_id, "ada*")
+        or fnmatch.fnmatch(model_id, "v3*")
+        or fnmatch.fnmatch(model_id, "*embedding*")
+    ):
+        return "openai"
+
+    if is_anthropic_model(model_id):
+        return "anthropic"
+
+    if fnmatch.fnmatch(model_id, GEMINI_PATTERN):
+        return "google"
+
+    return "unknown"