polyllmkit 0.1.1__tar.gz → 0.1.2__tar.gz

{polyllmkit-0.1.1 → polyllmkit-0.1.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: polyllmkit
-Version: 0.1.1
+Version: 0.1.2
 Summary: Streaming, multi-provider LLM bridge + markdown rendering toolkit for terminal apps.
 Requires-Dist: openai>=1.0 ; extra == 'all'
 Requires-Dist: claude-agent-sdk>=0.2 ; extra == 'all'

polyllmkit-0.1.2/pyproject.toml

@@ -0,0 +1,37 @@
+[project]
+name = "polyllmkit"
+version = "0.1.2"
+description = "Streaming, multi-provider LLM bridge + markdown rendering toolkit for terminal apps."
+requires-python = ">=3.11"
+dependencies = []
+
+[project.optional-dependencies]
+bridge = ["openai>=1.0"]
+claude = ["claude-agent-sdk>=0.2"]
+anthropic = ["anthropic>=0.40"]
+google = ["google-genai>=1.0"]
+all = [
+    "openai>=1.0",
+    "claude-agent-sdk>=0.2",
+    "anthropic>=0.40",
+    "google-genai>=1.0",
+]
+md = [
+    "rich>=13.0",
+    "textual>=0.50",
+    "markdown-it-py>=3.0",
+]
+
+[build-system]
+requires = ["uv_build>=0.5,<0.12"]
+build-backend = "uv_build"
+
+[tool.uv.build-backend]
+module-root = "src"
+module-name = "llmkit"
+
+[dependency-groups]
+dev = [
+    "ruff>=0.6",
+    "mypy>=1.10",
+]

polyllmkit-0.1.1/pyproject.toml → polyllmkit-0.1.2/pyproject.toml.orig

@@ -4,7 +4,7 @@
 # module-name below): consumers keep `import llmkit`, `pip install polyllmkit`.
 [project]
 name = "polyllmkit"
-version = "0.1.1"
+version = "0.1.2"
 description = "Streaming, multi-provider LLM bridge + markdown rendering toolkit for terminal apps."
 requires-python = ">=3.11"  # tomllib (bridge.config) is stdlib only from 3.11
 # Base install pulls nothing — pick the subpackage(s) you need via extras.

{polyllmkit-0.1.1 → polyllmkit-0.1.2}/src/llmkit/bridge/adapters/claude_code.py

@@ -58,15 +58,15 @@ def _thinking_budget() -> int:
 
 def _build_options(sdk: Any, provider: Provider, request: ChatRequest) -> Any:
     """Assemble ClaudeAgentOptions for a tool-free single-turn chat."""
-    opts: dict[str, Any] = dict(
-        tools=[],  # no built-in tools — plain chat, no file/shell access
-        allowed_tools=[],
-        disallowed_tools=[],
-        max_turns=1,
-        permission_mode="bypassPermissions",
-        include_partial_messages=True,  # token-level streaming deltas
-        setting_sources=[],  # ignore project/user CLAUDE.md, settings, etc.
-    )
+    opts: dict[str, Any] = {
+        "tools": [],  # no built-in tools — plain chat, no file/shell access
+        "allowed_tools": [],
+        "disallowed_tools": [],
+        "max_turns": 1,
+        "permission_mode": "bypassPermissions",
+        "include_partial_messages": True,  # token-level streaming deltas
+        "setting_sources": [],  # ignore project/user CLAUDE.md, settings, etc.
+    }
     if provider.model:
         opts["model"] = provider.model
     if request.system:
@@ -87,13 +87,22 @@ def _build_options(sdk: Any, provider: Provider, request: ChatRequest) -> Any:
         opts["thinking"] = {"type": "disabled"}
         return sdk.ClaudeAgentOptions(**opts)
 
+    # ``display: "summarized"`` is required on current models: they default it
+    # to ``omitted``, which silences thinking entirely — no deltas stream and
+    # the final ThinkingBlock is empty (verified: enabled+budget without display
+    # streams 0 chars of thinking; with display, ~tokens arrive). Mirrors the
+    # anthropic adapter's same-named field.
     flag = provider.enable_thinking or "auto"
     if flag == "false":
         opts["thinking"] = {"type": "disabled"}
     elif flag == "true":
-        opts["thinking"] = {"type": "enabled", "budget_tokens": _thinking_budget()}
+        opts["thinking"] = {
+            "type": "enabled",
+            "budget_tokens": _thinking_budget(),
+            "display": "summarized",
+        }
     else:
-        opts["thinking"] = {"type": "adaptive"}
+        opts["thinking"] = {"type": "adaptive", "display": "summarized"}
 
     return sdk.ClaudeAgentOptions(**opts)
 
@@ -146,7 +155,7 @@ async def _run(
 
 def structured_claude_code(
     provider: Provider, request: ChatRequest
-) -> "tuple[int, Any]":
+) -> tuple[int, Any]:
     """Structured output via the SDK's native ``output_format`` (set in
     ``_build_options`` when the request carries a schema). Runs the single-turn
     query and returns ``(exit_code, structured_output)`` from the ResultMessage —

{polyllmkit-0.1.1 → polyllmkit-0.1.2}/src/llmkit/bridge/bridge.py

@@ -336,7 +336,7 @@ def complete(
 ) -> int:
     """Run a text-completion (FIM) request, streaming text to ``out``
     (default stdout). ``openai-compatible`` only."""
-    out = out if out is not None else sys.stdout
+    sink = out if out is not None else sys.stdout
     provider = with_defaults(provider)
     if not provider_supports_complete(provider):
         print(
@@ -349,12 +349,12 @@ def complete(
         return 2
 
     client = build_client(provider)
-    kwargs: dict[str, Any] = dict(
-        model=provider.model,
-        prompt=request.prompt,
-        max_tokens=provider.max_tokens,
-        temperature=provider.temperature,
-    )
+    kwargs: dict[str, Any] = {
+        "model": provider.model,
+        "prompt": request.prompt,
+        "max_tokens": provider.max_tokens,
+        "temperature": provider.temperature,
+    }
     if request.suffix:
         kwargs["suffix"] = request.suffix
     if request.stop:
@@ -364,8 +364,8 @@ def complete(
         for chunk in client.completions.create(**kwargs):
             t = chunk.choices[0].text if chunk.choices else ""
             if t:
-                out.write(t)
-                out.flush()
+                sink.write(t)
+                sink.flush()
     except KeyboardInterrupt:
         return 130
     except Exception as e: