From 6257c898001849251eadbe630e2258623e8e7e0c Mon Sep 17 00:00:00 2001
From: inference-bot <inference-bot@hermes>
Date: Sat, 5 Sep 2026 22:24:43 -0600
Subject: [PATCH] Make key creation idempotent (reuse existing key by alias)

---
 app/main.py | 37 +++++++++++++++++++++++++++++++++++--
 1 file changed, 35 insertions(+), 2 deletions(-)

diff --git a/app/main.py b/app/main.py
index 13f2e97..95ea98a 100644
--- a/app/main.py
+++ b/app/main.py
@@ -222,14 +222,47 @@ async def cloudron_send_invite(user_id: str, email: str) -> None:
         r.raise_for_status()
 
 
+async def litellm_find_key_by_alias(alias: str) -> str | None:
+    """Find an existing key's token by its alias (key/list returns tokens)."""
+    async with httpx.AsyncClient() as client:
+        r = await client.get(
+            f"{LITELLM_BASE}/key/list",
+            headers={"Authorization": f"Bearer {LITELLM_MASTER_KEY}"},
+        )
+        r.raise_for_status()
+        tokens = r.json().get("keys", [])
+        for token in tokens:
+            info = await client.get(
+                f"{LITELLM_BASE}/key/info",
+                headers={"Authorization": f"Bearer {LITELLM_MASTER_KEY}"},
+                params={"key": token},
+            )
+            if info.status_code == 200:
+                data = info.json().get("info", {})
+                if data.get("key_alias") == alias:
+                    return token
+    return None
+
+
 async def litellm_create_key(email: str, budget: float) -> str:
-    """Create a LiteLLM virtual key for a member with a budget cap."""
+    """Create a LiteLLM virtual key for a member with a budget cap.
+
+    Idempotent: if a key with this alias already exists, return its token.
+    """
+    alias = f"member:{email}"
+
+    # If the key already exists (e.g. from a prior partial attempt), reuse it.
+    existing = await litellm_find_key_by_alias(alias)
+    if existing:
+        logger.info("Reusing existing LiteLLM key for %s", email)
+        return existing
+
     async with httpx.AsyncClient() as client:
         r = await client.post(
             f"{LITELLM_BASE}/key/generate",
             headers={"Authorization": f"Bearer {LITELLM_MASTER_KEY}"},
             json={
-                "key_alias": f"member:{email}",
+                "key_alias": alias,
                 "max_budget": budget,
                 "budget_duration": "30d",
                 "models": ["deepseek-v4-flash", "gpt-oss-120b"],