Troubleshooting¶
The short version of every support conversation we've had.
Morning Report¶
Not arriving at all 1. Team → Features → Morning Report — is the toggle on? (Admins only.) 2. Plan: the report requires Growth or above (active trials qualify). 3. One report per organization per day — a manual "send now" earlier in the day counts as that day's report.
Arriving at the wrong time — set your Timezone first, then the send hour. The default is 08:00 UTC until you change it.
In email but not Slack — Slack delivery needs a connected workspace (Integrations → Slack) and a channel the DXDT app can post to. If the configured channel is private, invite the DXDT app to it.
Approve/Dismiss link says expired — decision links live for 7 days. The recommendation is still available on the Pulse page. "Already decided" means someone else got there first — that's by design.
Teaching & memory¶
I taught a fact but the agent doesn't use it — check it's active on
the Memory page (retired facts don't inject). The agent injects the 50
most recent facts; very large memories favor the newest.
The teach button saved a "duplicate" — an equivalent fact (same normalized text) already exists. Find it on the Memory page; if it was retired, teaching it again reactivated it.
Plan review¶
"not in your mapped graph — no blast-radius data" — DXDT has no graph coverage for that resource. Connect the relevant integration; coverage grows on the next mapping run.
Exit code 1 in CI — the overall risk met your --fail-on threshold.
Read the annotations (each cites its evidence); to observe without
blocking, use --fail-on never.
401/403 from /api/v1/plan-review — the PAT is missing, revoked, or
malformed. dxdt login to mint a fresh one; CI needs the token in
DXDT_TOKEN.
Findings & savings¶
"Why did my identified savings drop?" — duplicate clustering. When multiple findings cover the same resources, DXDT counts each resource once. The lower number is the defensible one.
A finding keeps coming back after I dismiss it — dismissing records your decision on that finding; if the underlying condition persists, a genuinely new detection can still appear. If the condition is expected, teach it as a fact — that's what suppresses it upstream.
A finding resolved "by itself" — the reinvestigation sweep re-checks stale findings and closes them when it can prove the issue is gone. The evidence and the resolver are on the finding in the Resolved tab; if the sweep got it wrong, Reopen puts it back.
The resolution wizard's Verify keeps saying "not visible yet" — cloud APIs can lag a real fix by minutes. If it persists past a few retries, the fix may not have taken effect where the check is looking — open the step's verification command to see exactly what's being checked. See Resolution wizard for all wizard states.
General¶
402 with upgrade_required — the feature is above your plan tier.
CLI exit codes — 0 ok, 1 error, 2 auth, 3 quota/rate-limit.
Long agent runs seem stuck — long turns keep executing server-side;
dxdt attach rejoins the stream.
Still stuck? Every page in the console links to support, and
security@dxdt.ai reaches us for anything sensitive.