From e33e720b26356a794d869c9efbfee05a514c063d Mon Sep 17 00:00:00 2001
From: Cody <cody@dkdc.dev>
Date: Fri, 3 Apr 2026 23:06:31 -0400
Subject: [PATCH] feat(py): add auth params to chat() and create Python tests

Expose auth_bearer, auth_header, auth_value in the high-level chat()
wrapper to match _chat_raw(). Add test_api.py with unit tests and e2e
tests (marked, skipped by default).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---
 py/stringflow/__init__.py | 14 +++++-
 py/stringflow/test_api.py | 92 +++++++++++++++++++++++++++++++++++++++
 pyproject.toml            |  4 ++
 3 files changed, 109 insertions(+), 1 deletion(-)
 create mode 100644 py/stringflow/test_api.py

diff --git a/py/stringflow/__init__.py b/py/stringflow/__init__.py
index 412e968..a140edf 100644
--- a/py/stringflow/__init__.py
+++ b/py/stringflow/__init__.py
@@ -16,6 +16,9 @@ def chat(
     wire_format: str = "messages",
     model: str | None = None,
     max_tokens: int | None = None,
+    auth_bearer: str | None = None,
+    auth_header: str | None = None,
+    auth_value: str | None = None,
 ) -> list[Message]:
     """Chat with an LLM. Returns conversation history you can pass back in.
 
@@ -35,7 +38,16 @@ def chat(
         )
 
     try:
-        response = _chat_raw(base_url, messages, wire_format, model, max_tokens)
+        response = _chat_raw(
+            base_url,
+            messages,
+            wire_format,
+            model,
+            max_tokens,
+            auth_bearer,
+            auth_header,
+            auth_value,
+        )
     except ConnectionError as e:
         raise ConnectionError(
             f"cannot reach LLM server at {base_url} — is dkdc-ai running?\n"
diff --git a/py/stringflow/test_api.py b/py/stringflow/test_api.py
new file mode 100644
index 0000000..1992bf1
--- /dev/null
+++ b/py/stringflow/test_api.py
@@ -0,0 +1,92 @@
+"""Tests for the stringflow Python API.
+
+E2E tests require a running llama-server on localhost:8080.
+Run with: uv run pytest py/stringflow/test_api.py
+"""
+
+import pytest
+
+import stringflow as sf
+
+
+# ============================================================================
+# Unit tests (no server required)
+# ============================================================================
+
+
+class TestChatInput:
+    def test_string_message_builds_history(self):
+        """chat() should accept a string and build a user message tuple."""
+        # We can't call chat() without a server, but we can test the TypeError path
+        with pytest.raises(TypeError):
+            sf.chat(42)  # type: ignore
+
+    def test_rejects_invalid_type(self):
+        with pytest.raises(TypeError, match="must be str or list"):
+            sf.chat(123)  # type: ignore
+
+    def test_connection_error_without_server(self):
+        """chat() should raise ConnectionError when no server is running."""
+        with pytest.raises((ConnectionError, Exception)):
+            sf.chat("hi", base_url="http://localhost:19999")
+
+
+class TestDefaults:
+    def test_default_url(self):
+        assert sf.DEFAULT_URL == "http://localhost:8080"
+
+    def test_exports(self):
+        assert hasattr(sf, "chat")
+        assert hasattr(sf, "health_check")
+        assert hasattr(sf, "DEFAULT_URL")
+        assert hasattr(sf, "Message")
+
+
+# ============================================================================
+# E2E tests (require running llama-server on localhost:8080)
+# ============================================================================
+
+
+@pytest.mark.e2e
+class TestChatE2E:
+    def test_simple_chat(self):
+        result = sf.chat("Reply with exactly the word 'pong' and nothing else.")
+        assert isinstance(result, list)
+        assert len(result) == 2
+        assert result[0] == (
+            "user",
+            "Reply with exactly the word 'pong' and nothing else.",
+        )
+        assert result[1][0] == "assistant"
+        assert len(result[1][1]) > 0
+
+    def test_multi_turn(self):
+        r1 = sf.chat("My name is TestBot.")
+        assert len(r1) == 2
+        r2 = sf.chat("What is my name?", r1)
+        assert len(r2) == 4
+        assert r2[2] == ("user", "What is my name?")
+        assert r2[3][0] == "assistant"
+
+    def test_message_list_input(self):
+        messages = [("user", "Reply with exactly 'hello'.")]
+        result = sf.chat(messages)
+        assert len(result) == 2
+        assert result[1][0] == "assistant"
+
+    def test_wire_format_completions(self):
+        result = sf.chat("Say hi.", wire_format="completions")
+        assert len(result) == 2
+        assert result[1][0] == "assistant"
+
+    def test_wire_format_responses(self):
+        result = sf.chat("Say hi.", wire_format="responses")
+        assert len(result) == 2
+        assert result[1][0] == "assistant"
+
+
+@pytest.mark.e2e
+class TestHealthCheckE2E:
+    def test_health_check(self):
+        result = sf.health_check()
+        assert isinstance(result, str)
diff --git a/pyproject.toml b/pyproject.toml
index a4c266e..93e81c8 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -24,6 +24,10 @@ python-packages = ["stringflow"]
 python-source = "py"
 manifest-path = "crates/stringflow-py/Cargo.toml"
 
+[tool.pytest.ini_options]
+markers = ["e2e: end-to-end tests requiring a running llama-server"]
+addopts = "-m 'not e2e'"
+
 [build-system]
 requires = ["maturin>=1.0,<2.0"]
 build-backend = "maturin"