feat(analytics): report to collect.mindshub.ai and send an identifying agent (ENG-1355)

anton/analytics.py

@@ -8,16 +8,29 @@
 Where these events actually land
 ================================
 
-The collector lambda RELAYS into PostHog — verified 2026-08-06 against
-project 355390 ("Anton"), where anton's events appear tagged
-``source = mindsdb-zoominfo-lambda``.  Two consequences worth knowing before
-adding a caller:
-
-* **Every ``extra`` kwarg becomes a queryable PostHog event property.**  Not
-  an allowlist — ``ds_connect_success`` carries its ``engine=postgres``
-  through, and ``turn_completed`` carries its full token breakdown.  So the
-  parameter names here ARE the analytics schema; renaming one silently breaks
-  whatever queries or dashboards read it.
+The collector relays into PostHog project 355390 ("Anton"), where these events
+appear tagged ``source = mindshub-analytics-collector``.  Rows written before
+2026-08-11 carry ``source = mindsdb-zoominfo-lambda`` instead: that was a
+different collector, in a different AWS account, and installs old enough to
+still hold the previous ``analytics_url`` keep reporting to it.  A query
+covering both needs to accept either value.
+
+Consequences worth knowing before adding a caller:
+
+* **Every ``extra`` kwarg becomes a queryable PostHog event property**, apart
+  from a small denylist of credential- and address-shaped names, and any value
+  that looks like an email address.  ``ds_connect_success`` carries its
+  ``engine=postgres`` through and ``turn_completed`` carries its full token
+  breakdown, with no collector change needed for either.  So the parameter
+  names here ARE the analytics schema; renaming one silently breaks whatever
+  queries or dashboards read it.
+* **This was not true until 2026-08-11.**  The previous collector copied a
+  fixed list of five property names and relayed only actions starting with
+  a lowercase ``anton_`` or ``ds_connect_``, discarding everything else and
+  answering HTTP 200 regardless.  ``turn_completed`` and its 27 properties
+  therefore went nowhere at all from the day they shipped.  If a property
+  seems to be missing, check what the collector accepts before assuming the
+  caller is at fault.
 * **``distinct_id`` is the ``aid`` fingerprint, so these events are
   per-INSTALL, not per-user.**  They do not join to the Keycloak ``sub`` that
   the console, the desktop renderer's PostHog client, and the billing mirror
@@ -82,6 +95,14 @@
 
 _TIMEOUT = 3  # seconds
 
+# Sent instead of urllib's default ``Python-urllib/3.x``, which Cloudflare's bot
+# protection answers with 403 on the mindshub.ai zone.  The collector host is
+# deliberately not proxied, so this is belt and braces rather than the only
+# guard, but ``_fire`` discards its response either way: anything that blocks
+# this request disappears without a trace, so it is worth not looking like a
+# script.
+_USER_AGENT = "anton-analytics/1.0"
+
 # Cached after first computation — the fingerprint never changes within
 # a process, so computing it once is sufficient.
 _cached_aid: str | None = None
@@ -194,6 +215,7 @@ def send_event(settings: "AntonSettings", action: str, **extra: str) -> None:
 def _fire(url: str) -> None:
     """Perform the actual HTTP GET.  Runs inside a daemon thread."""
     try:
-        urllib.request.urlopen(url, timeout=_TIMEOUT)
+        req = urllib.request.Request(url, headers={"User-Agent": _USER_AGENT})
+        urllib.request.urlopen(req, timeout=_TIMEOUT)
     except Exception:
         pass

anton/config/settings.py

@@ -136,7 +136,9 @@ def _map_minds_cloud_to_openai_compatible(cls, v: object) -> object:
 
     # Analytics — anonymous usage events (set ANTON_ANALYTICS_ENABLED=false to opt out)
     analytics_enabled: bool = True
-    analytics_url: str = "https://x6nik28qi6.execute-api.us-east-2.amazonaws.com/default/zoomInfoCollector"
+    # A hostname we own rather than an API Gateway id, so the collector can move
+    # without a release to follow it. Override with ANTON_ANALYTICS_URL.
+    analytics_url: str = "https://collect.mindshub.ai/collect"
 
     # Minds datasource integration
     minds_enabled: bool = True  # use Minds server as LLM provider

tests/test_analytics.py

@@ -103,3 +103,37 @@ def test_send_event_sends_when_not_ci(monkeypatch):
     assert params["action"] == "anton_query"
     assert params["llm_provider"] == "openai"
     assert "is_ci" not in params  # flag removed; CI events aren't sent at all
+
+
+def test_fire_sends_an_identifying_user_agent(monkeypatch):
+    """urllib's default ``Python-urllib/3.x`` agent is answered with 403 by the
+    bot rules on the mindshub.ai zone, and ``_fire`` throws its response away,
+    so an event blocked for looking like a script would vanish with no trace.
+    """
+    seen: list[object] = []
+
+    def _fake_urlopen(request, timeout=None):
+        seen.append(request)
+        return None
+
+    monkeypatch.setattr(analytics.urllib.request, "urlopen", _fake_urlopen)
+
+    analytics._fire("https://example.test/collect?action=anton_query")
+
+    assert len(seen) == 1
+    agent = seen[0].get_header("User-agent")
+    assert agent == analytics._USER_AGENT
+    assert "python-urllib" not in agent.lower()
+
+
+def test_default_collector_url_is_a_host_we_control():
+    """Guards the regression this replaced: the previous default was a raw
+    ``*.execute-api.amazonaws.com`` id in an account nobody could deploy to, so
+    the endpoint could not be changed without shipping a new release to every
+    install.
+    """
+    from anton.config.settings import AntonSettings
+
+    url = AntonSettings.model_fields["analytics_url"].default
+    assert url.startswith("https://collect.")
+    assert "execute-api" not in url