Use direct OpenAI defaults for matchmaking

.env.example

@@ -14,6 +14,12 @@ REDIS_URL=redis://localhost:6379/0
 # Inputs / engine (wired in later slices)
 APIFY_TOKEN=
 OPENAI_API_KEY=
+OPENAI_API_BASE=
+DSPY_API_BASE=
+DSPY_API_KEY=
+CURATE_MODEL=gpt-4o-mini
+SUGGEST_MODEL=gpt-4o-mini
+EMBED_MODEL=text-embedding-3-small
 
 # QScore (consumed)
 QSCORE_BASE_URL=http://localhost:8004

app/config.py

@@ -33,12 +33,12 @@ class Settings(BaseSettings):
     OPENAI_API_KEY: str | None = None
     OPENAI_API_BASE: str | None = None
 
-    # LLM gateway (OpenAI-compatible, e.g. opencode/OpenRouter) for chat curation.
+    # LLM gateway override for chat curation. Leave blank to use OPENAI_API_KEY directly.
     DSPY_API_BASE: str | None = None
     DSPY_API_KEY: str | None = None
-    CURATE_MODEL: str = "claude-opus-4-8"          # the senior recruiter (Opus, latest) — Opus's call is final
-    SUGGEST_MODEL: str = "claude-haiku-4-5"        # the bubbling engine — fast + cheap (latency-sensitive UI)
-    EMBED_MODEL: str = "text-embedding-3-small"    # the vibe-engineer (direct OpenAI or OpenAI-compatible embeddings API)
+    CURATE_MODEL: str = "gpt-4o-mini"             # the senior recruiter
+    SUGGEST_MODEL: str = "gpt-4o-mini"            # the bubbling engine
+    EMBED_MODEL: str = "text-embedding-3-small"   # direct OpenAI embeddings
     ENGINE_LLM_ENABLED: bool = True                # off / key absent → white-box sift + templated cards (safety net)
     SIFT_TOP_K: int = 18                           # candidate pool the curator reads
 

app/engine/llm.py

@@ -1,7 +1,7 @@
 """LLM clients — ONE place to build them.
 
 Two backends, by capability:
-- **Gateway** (OpenAI-compatible, e.g. opencode/OpenRouter) → chat for the curator.
+- **Gateway** → chat for the curator, using DSPY_* when configured or direct OpenAI otherwise.
 - **Embeddings** → direct OpenAI or an OpenAI-compatible embeddings API via OPENAI_API_BASE.
 
 Both keys come from config/.env (git-ignored). Each builder returns `None` when unconfigured so the
@@ -16,12 +16,16 @@ from app.config import get_settings
 
 @lru_cache
 def gateway_client():
-    """OpenAI-compatible client for the opencode.ai/zen gateway — chat (Opus). None if unconfigured."""
+    """Chat client for curation/suggestions. Prefer DSPY_* override, otherwise use OPENAI_API_KEY."""
     s = get_settings()
-    if not (s.DSPY_API_BASE and s.DSPY_API_KEY):
+    api_key = s.DSPY_API_KEY or s.OPENAI_API_KEY
+    if not api_key:
         return None
     from openai import OpenAI
-    return OpenAI(base_url=s.DSPY_API_BASE, api_key=s.DSPY_API_KEY, timeout=60, max_retries=2)
+    kwargs = {"api_key": api_key, "timeout": 60, "max_retries": 2}
+    if s.DSPY_API_BASE:
+        kwargs["base_url"] = s.DSPY_API_BASE
+    return OpenAI(**kwargs)
 
 
 @lru_cache
@@ -38,6 +42,6 @@ def embed_client():
 
 
 def curate_enabled() -> bool:
-    """Is the Opus curator usable? (flag on + gateway key present)."""
+    """Is the curator usable? (flag on + any chat key present)."""
     s = get_settings()
-    return bool(s.ENGINE_LLM_ENABLED and s.DSPY_API_BASE and s.DSPY_API_KEY)
+    return bool(s.ENGINE_LLM_ENABLED and (s.DSPY_API_KEY or s.OPENAI_API_KEY))