Add production error monitoring via Sentry

.env.example

@@ -12,3 +12,8 @@ DAILY_CALL_LIMIT=50
 
 # Production frontend origin for CORS (e.g. https://your-app.railway.app)
 # ALLOWED_ORIGIN=https://your-app.railway.app
+
+# Sentry DSN for production error monitoring — optional, get one free at
+# https://sentry.io. Leave unset to disable error monitoring entirely (safe
+# default for local dev/CI).
+# SENTRY_DSN=https://...@o0.ingest.sentry.io/0

agent/main.py

@@ -5,6 +5,7 @@
 from datetime import date
 
 import anthropic
+import sentry_sdk
 from dotenv import load_dotenv
 
 from agent import db
@@ -249,6 +250,7 @@ def _run_tools(response_content: list, user_id: int, on_result=None, source: str
                 result = TOOL_HANDLERS[block.name](**kwargs)
             except Exception:
                 logger.error("tool %s(%s) failed:\n%s", block.name, kwargs, traceback.format_exc())
+                sentry_sdk.capture_exception()
                 result = {"status": "error", "message": f"{block.name} failed unexpectedly — tell the user and don't retry automatically."}
             db.record_usage(user_id, "tool", block.name, source)
             print(f"[tool] {block.name}({kwargs}) -> {result}")

api/server.py

@@ -7,6 +7,7 @@
 from datetime import date
 from pathlib import Path
 
+import sentry_sdk
 from fastapi import Depends, FastAPI, HTTPException
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.responses import FileResponse, StreamingResponse
@@ -37,6 +38,16 @@
 
 logger = logging.getLogger(__name__)
 
+# Error monitoring — a no-op until SENTRY_DSN is set (sentry_sdk.init(dsn=None)
+# disables the SDK entirely, so this is safe to leave unconfigured in dev/CI).
+# Error capture only, no performance tracing — that's a separate cost/scope
+# this app doesn't need yet. send_default_pii is explicitly off: the SDK
+# auto-instruments the anthropic client, and every Claude call here carries
+# real expense/income descriptions — that must never leave this app for a
+# third party, on top of the SYSTEM prompt/user messages being sensitive on
+# their own.
+sentry_sdk.init(dsn=os.environ.get("SENTRY_DSN"), traces_sample_rate=0.0, send_default_pii=False)
+
 DAILY_CALL_LIMIT = int(os.environ.get("DAILY_CALL_LIMIT", 50))
 
 
@@ -110,6 +121,7 @@ def generate():
                 yield f"data: {json.dumps(event)}\n\n"
         except Exception:
             logger.error("stream_chat error:\n%s", traceback.format_exc())
+            sentry_sdk.capture_exception()
             yield f"data: {json.dumps({'error': 'Something went wrong. Please try again.'})}\n\n"
         yield "data: [DONE]\n\n"
 

pyproject.toml

@@ -10,6 +10,7 @@ dependencies = [
     "psycopg2-binary>=2.9.12",
     "python-dotenv>=1.2.2",
     "python-jose[cryptography]>=3.5.0",
+    "sentry-sdk>=2.65.0",
     "uvicorn>=0.49.0",
 ]
 

tests/test_agent.py

@@ -125,6 +125,21 @@ def failing_handler(**kw):
     assert results[2]["content"] == str({"ok": True})
 
 
+def test_run_tools_reports_failing_handler_to_sentry(monkeypatch):
+    captured = []
+    monkeypatch.setattr(main.sentry_sdk, "capture_exception", lambda: captured.append(True))
+
+    def failing_handler(**kw):
+        raise ValueError("boom")
+
+    monkeypatch.setitem(main.TOOL_HANDLERS, "save_expense", failing_handler)
+    block = make_block("tool_use", name="save_expense", input={"amount": 5}, id="tool_1")
+
+    main._run_tools([block], user_id=1)
+
+    assert captured == [True]
+
+
 # --- _serialize_block --------------------------------------------------------
 # Required for persisting assistant turns to Postgres — a live Anthropic SDK
 # response block in production, but test doubles are plain dicts or

tests/test_api.py

@@ -302,6 +302,21 @@ def fake_stream_chat(message, user_id, username, images, source):
     assert "data: [DONE]" in response.text
 
 
+def test_chat_stream_endpoint_reports_exception_to_sentry(monkeypatch, auth_headers):
+    captured = []
+    monkeypatch.setattr(server.sentry_sdk, "capture_exception", lambda: captured.append(True))
+
+    def fake_stream_chat(message, user_id, username, images):
+        raise RuntimeError("boom")
+        yield  # pragma: no cover — makes this a generator, never reached
+
+    monkeypatch.setattr(server, "stream_chat", fake_stream_chat)
+
+    client.post("/chat/stream", json={"message": "hi"}, headers=auth_headers)
+
+    assert captured == [True]
+
+
 # --- check_rate_limit --------------------------------------------------
 
 def test_check_rate_limit_allows_calls_under_limit(user_id):