cade-cli 0.20.1__tar.gz → 0.20.2__tar.gz

{cade_cli-0.20.1 → cade_cli-0.20.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cade-cli
-Version: 0.20.1
+Version: 0.20.2
 Summary: Cade - The CLI Agent from Arcade.dev (install: curl -fsSL https://arcadeagent.dev/install | sh)
 Project-URL: Homepage, https://arcade.dev
 Project-URL: Documentation, https://arcadeagent.dev/docs

{cade_cli-0.20.1 → cade_cli-0.20.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "cade-cli"
-version = "0.20.1"
+version = "0.20.2"
 description = "Cade - The CLI Agent from Arcade.dev (install: curl -fsSL https://arcadeagent.dev/install | sh)"
 readme = "README.md"
 requires-python = ">=3.11"

{cade_cli-0.20.1 → cade_cli-0.20.2}/src/cadecoder/host_plugin/assets/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "cade",
   "description": "Your persistent on-device agent: notes memory, durable tasks, and connected tools via Ask.",
-  "version": "0.20.1",
+  "version": "0.20.2",
   "author": {
     "name": "Arcade.dev",
     "url": "https://docs.cade.dev"

{cade_cli-0.20.1 → cade_cli-0.20.2}/src/cadecoder/host_plugin/assets/.cursor-plugin/plugin.json

@@ -2,7 +2,7 @@
   "name": "cade",
   "displayName": "Cade",
   "description": "Your persistent on-device agent: notes memory, durable tasks, and connected tools via Ask.",
-  "version": "0.20.1",
+  "version": "0.20.2",
   "author": {
     "name": "Arcade.dev",
     "email": "support@arcade.dev"

{cade_cli-0.20.1 → cade_cli-0.20.2}/src/cadecoder/providers/http_client.py

@@ -1,9 +1,11 @@
 """Shared httpx client helpers for LLM providers.
 
 Frozen/PyInstaller builds have hit ``zlib.error: Error -3 while decompressing
-data: incorrect header check`` on Anthropic SSE streams when httpx negotiates
-brotli (``br``) or otherwise mishandles ``Content-Encoding``. Pinning
-``Accept-Encoding`` to gzip/deflate avoids that path while keeping compression.
+data: incorrect header check`` on Anthropic/OpenAI-compatible SSE streams when
+httpx negotiates compression and the body is not actually compressed — common
+with local vLLM/Ollama and some proxies. Default ``Accept-Encoding: identity``
+so the client never attempts to decompress streams. Override with
+``CADE_HTTP_ACCEPT_ENCODING`` (e.g. ``gzip, deflate``) if you want compression.
 
 Also: shell profiles often export ``SSL_CERT_FILE`` / ``REQUESTS_CA_BUNDLE`` to a
 path that exists on one machine and not another. httpx then raises
@@ -20,9 +22,10 @@ from pathlib import Path
 
 import httpx
 
-# Do not advertise ``br`` / ``zstd`` — those decoders are fragile in frozen
-# binaries and have produced "incorrect header check" failures on streams.
-_SAFE_ACCEPT_ENCODING = "gzip, deflate"
+# Never advertise ``br`` / ``zstd`` / ``gzip`` by default — those decoders are
+# fragile in frozen binaries and with OpenAI-compatible SSE servers that omit
+# or mis-set Content-Encoding ("incorrect header check").
+_DEFAULT_ACCEPT_ENCODING = "identity"
 
 _SSL_ENV_KEYS = ("SSL_CERT_FILE", "REQUESTS_CA_BUNDLE", "CURL_CA_BUNDLE")
 
@@ -30,6 +33,12 @@ _SSL_ENV_KEYS = ("SSL_CERT_FILE", "REQUESTS_CA_BUNDLE", "CURL_CA_BUNDLE")
 CLEARED_SSL_ENV_KEYS: list[str] = []
 
 
+def accept_encoding_header() -> str:
+    """Return the Accept-Encoding value for provider HTTP clients."""
+    raw = os.environ.get("CADE_HTTP_ACCEPT_ENCODING", "").strip()
+    return raw or _DEFAULT_ACCEPT_ENCODING
+
+
 def scrub_broken_ssl_env() -> list[str]:
     """Unset SSL CA env vars that point at missing files.
 
@@ -94,7 +103,7 @@ def make_async_http_client(*, timeout: float = 600.0) -> httpx.AsyncClient:
     """Return an AsyncClient safe for Anthropic/OpenAI streaming in frozen apps."""
     prepare_ssl_env()
     return httpx.AsyncClient(
-        headers={"Accept-Encoding": _SAFE_ACCEPT_ENCODING},
+        headers={"Accept-Encoding": accept_encoding_header()},
         timeout=httpx.Timeout(timeout, connect=10.0),
         follow_redirects=True,
         verify=ssl_verify(),
@@ -105,7 +114,7 @@ def make_sync_http_client(*, timeout: float = 60.0) -> httpx.Client:
     """Return a sync Client with the same encoding constraints."""
     prepare_ssl_env()
     return httpx.Client(
-        headers={"Accept-Encoding": _SAFE_ACCEPT_ENCODING},
+        headers={"Accept-Encoding": accept_encoding_header()},
         timeout=httpx.Timeout(timeout, connect=10.0),
         follow_redirects=True,
         verify=ssl_verify(),
@@ -126,6 +135,7 @@ def is_decompress_error(exc: BaseException) -> bool:
 
 __all__ = [
     "CLEARED_SSL_ENV_KEYS",
+    "accept_encoding_header",
     "is_decompress_error",
     "make_async_http_client",
     "make_sync_http_client",

{cade_cli-0.20.1 → cade_cli-0.20.2}/src/cadecoder/providers/openai.py

@@ -220,12 +220,18 @@ class OpenAIProvider(Provider):
         # Ask OpenAI (and compatible servers that honor it) for a final usage
         # chunk so OTEL spans can record prompt/completion tokens. Some local
         # OpenAI-compatible servers reject stream_options — fall back cleanly.
-        want_usage = os.environ.get("CADE_OTEL", "").strip().lower() in (
-            "1",
-            "true",
-            "yes",
-            "on",
-        )
+        want_usage = False
+        if os.environ.get("CADE_OTEL", "").strip().lower() in ("1", "true", "yes", "on"):
+            # Official OpenAI honors stream_options; many local OpenAI-compatible
+            # servers (vLLM, etc.) mishandle it. Opt in for custom base_url with
+            # CADE_OTEL_STREAM_USAGE=1.
+            force_usage = os.environ.get("CADE_OTEL_STREAM_USAGE", "").strip().lower() in (
+                "1",
+                "true",
+                "yes",
+                "on",
+            )
+            want_usage = force_usage or not self.base_url
         if want_usage:
             openai_request["stream_options"] = {"include_usage": True}