verbose live UI + tool-level SSE events + Groq default + regression tests

api/llm.py

@@ -0,0 +1,91 @@
+"""LLM client — single `chat()` entry point that dispatches to the active provider.
+
+Provider selection is via `settings.llm_provider`. Supported:
+- "groq"      — hits Groq via the openai SDK (groq_base_url + groq_api_key).
+- "anthropic" — native Anthropic SDK.
+- "openai"    — openai SDK with the standard base_url.
+
+The Anthropic and OpenAI/Groq clients are cached so HTTP connection pools
+are reused across calls.
+"""
+
+from __future__ import annotations
+
+from functools import lru_cache
+from typing import Callable
+
+from anthropic import Anthropic
+from openai import OpenAI
+
+from api.config import get_settings
+
+
+# ── Provider clients (cached) ──
+
+@lru_cache(maxsize=1)
+def _anthropic() -> Anthropic:
+    return Anthropic(api_key=get_settings().anthropic_api_key)
+
+
+@lru_cache(maxsize=1)
+def _groq() -> OpenAI:
+    s = get_settings()
+    return OpenAI(api_key=s.groq_api_key, base_url=s.groq_base_url)
+
+
+@lru_cache(maxsize=1)
+def _openai() -> OpenAI:
+    s = get_settings()
+    return OpenAI(api_key=s.openai_api_key, base_url=s.openai_base_url)
+
+
+# ── Per-provider chat impls ──
+
+def _chat_anthropic(system: str, user: str, max_tokens: int) -> str:
+    s = get_settings()
+    msg = _anthropic().messages.create(
+        model=s.anthropic_model,
+        max_tokens=max_tokens,
+        system=system,
+        messages=[{"role": "user", "content": user}],
+    )
+    return "".join(b.text for b in msg.content if getattr(b, "type", None) == "text").strip()
+
+
+def _chat_openai_compat(client: OpenAI, model: str, system: str, user: str, max_tokens: int) -> str:
+    resp = client.chat.completions.create(
+        model=model,
+        max_tokens=max_tokens,
+        messages=[
+            {"role": "system", "content": system},
+            {"role": "user", "content": user},
+        ],
+    )
+    return (resp.choices[0].message.content or "").strip()
+
+
+def _chat_groq(system: str, user: str, max_tokens: int) -> str:
+    return _chat_openai_compat(_groq(), get_settings().groq_model, system, user, max_tokens)
+
+
+def _chat_openai(system: str, user: str, max_tokens: int) -> str:
+    return _chat_openai_compat(_openai(), get_settings().openai_model, system, user, max_tokens)
+
+
+_PROVIDERS: dict[str, Callable[[str, str, int], str]] = {
+    "groq": _chat_groq,
+    "anthropic": _chat_anthropic,
+    "openai": _chat_openai,
+}
+
+
+# ── Public surface ──
+
+def chat(*, system: str, user: str, max_tokens: int = 1024) -> str:
+    provider = get_settings().llm_provider.lower()
+    impl = _PROVIDERS.get(provider)
+    if impl is None:
+        raise ValueError(
+            f"unknown llm_provider {provider!r}; supported: {sorted(_PROVIDERS)}"
+        )
+    return impl(system, user, max_tokens)