From 0453cee01933be9bcc67a6862e97a394b0a43c13 Mon Sep 17 00:00:00 2001
From: inference-bot <inference-bot@hermes>
Date: Mon, 14 Sep 2026 14:18:59 -0600
Subject: [PATCH] Broker usage: add per-model token/spend breakdown +
 total_tokens + email

---
 app/main.py | 63 +++++++++++++++++++++++++++++++++++++++++++----------
 1 file changed, 52 insertions(+), 11 deletions(-)

diff --git a/app/main.py b/app/main.py
index a8571c5..20ab966 100644
--- a/app/main.py
+++ b/app/main.py
@@ -1183,14 +1183,19 @@ async def broker_revoke(name: str, request: Request):
 
 
 async def broker_get_usage(email: str) -> dict:
-    """Return the member's spend vs. balance, from their team.
+    """Return the member's spend vs. balance, plus per-model and token detail.
 
-    LiteLLM's team object exposes `spend` and `max_budget` directly, so we read
-    the member's team and return {balance, spend, remaining}. This is the
-    read-only primitive the member dashboard uses (no master key on the client).
+    Reads the member's team (spend + max_budget) and aggregates their recent
+    spend logs by model for token counts. This is the read-only primitive the
+    member dashboard uses (no master key on the client).
     """
     team_id = await litellm_get_or_create_team(email, get_member_balance(email))
+
+    balance = get_member_balance(email)
+    spend = 0.0
+
     async with httpx.AsyncClient() as client:
+        # 1. Team spend + budget
         r = await client.get(
             f"{LITELLM_BASE}/team/list",
             headers={"Authorization": f"Bearer {LITELLM_MASTER_KEY}"},
@@ -1198,14 +1203,50 @@ async def broker_get_usage(email: str) -> dict:
         r.raise_for_status()
         for t in r.json():
             if t.get("team_id") == team_id:
-                balance = float(t.get("max_budget") or get_member_balance(email))
+                balance = float(t.get("max_budget") or balance)
                 spend = float(t.get("spend") or 0.0)
-                return {
-                    "balance": balance,
-                    "spend": spend,
-                    "remaining": max(balance - spend, 0.0),
-                }
-    return {"balance": get_member_balance(email), "spend": 0.0, "remaining": get_member_balance(email)}
+                break
+
+        # 2. Per-model token/spend breakdown from spend logs (filtered by team).
+        # /spend/logs does not reliably filter by team_id server-side, so we
+        # filter client-side on the returned rows.
+        models = {}
+        total_tokens = 0
+        try:
+            r2 = await client.get(
+                f"{LITELLM_BASE}/spend/logs",
+                headers={"Authorization": f"Bearer {LITELLM_MASTER_KEY}"},
+            )
+            r2.raise_for_status()
+            rows = r2.json()
+            if isinstance(rows, dict):
+                rows = rows.get("data", rows.get("logs", []))
+            for row in rows if isinstance(rows, list) else []:
+                if row.get("team_id") != team_id:
+                    continue
+                model = row.get("model") or "unknown"
+                m = models.setdefault(model, {"spend": 0.0, "tokens": 0, "calls": 0})
+                m["spend"] += float(row.get("spend") or 0.0)
+                m["tokens"] += int(row.get("total_tokens") or 0)
+                m["calls"] += 1
+                total_tokens += int(row.get("total_tokens") or 0)
+        except Exception as e:  # spend logs are best-effort; don't fail the whole view
+            logger.warning("spend/logs fetch failed: %s", e)
+
+    # Sort models by spend descending (which models are draining the most).
+    model_list = [
+        {"model": name, **stats}
+        for name, stats in sorted(models.items(), key=lambda kv: kv[1]["spend"], reverse=True)
+    ]
+
+    return {
+        "email": email,
+        "balance": balance,
+        "spend": spend,
+        "remaining": max(balance - spend, 0.0),
+        "total_tokens": total_tokens,
+        "models": model_list,
+    }
 
 
 @app.get("/broker/usage")