Troubleshooting

A v1 doc on an early-access product. You will hit rough edges. Here are the ones we already know about and how to get past them.

The welcome email landed in spam

Outlook, Live, and Gmail are the usual suspects. Check the spam folder, mark the sender (update@etiakorp.com) as legitimate, and move the email to your inbox. Any future system emails will land in the right place.

I received several invite emails

This happens when provisioning retries. Use the most recent one. Older links may still work, but the latest is the canonical one.

The app URL doesn't load, or I see a 502 / 503

New tenant subdomains need 1–2 minutes for DNS and TLS to propagate. Wait, then refresh. If it's been more than 5 minutes, the cluster may be mid-sync — try again in another minute or two. If it's still not working after 10 minutes, see Getting help.

The activation link says "expired"

Action tokens in the activation email expire after a short window (usually a few hours). The fix is a fresh email — file a support case (with your slug) and we'll re-trigger the invite. If you already got several invite emails, try a more recent one before contacting support.

The onboarding wizard hangs at "Decomposing into tasks…"

Refresh the page. The wizard re-reads state from the server and you can pick up where you left off — your workspace choice is preserved. If the hang comes back after the refresh, contact support; this usually means something is mis-configured on the back end and we need to look.

A provider API key was rejected

Check the prefix:

• Anthropic keys start with sk-ant-…
• OpenAI (and compatible providers) keys start with sk-…
• Synthetic.new keys start with syn-…

If the prefix is wrong, you probably copied the wrong field from the provider's console. Regenerate from the provider's dashboard if the key is expired. The error message in Genesis may surface the raw HTTP status in older paths — 401 is "invalid key", 403 is "no access", 429 is "rate limited or no credits".

Mission stuck in "Decomposing" longer than 2 minutes

• Refresh the page.
• If your provider had a recent outage, the planner may be retrying; wait another minute.
• If 5 minutes elapse with no progress, cancel the mission and re-create the conversation.

Mission stuck in "Running" with no apparent progress

• Open the mission. Look at individual task status.
• Tasks in Pending beyond a few minutes → fleet may be saturated. Check the Tools tab of LLM Config: the fleet's Max Concurrent, each variant's Max Instances, and a live count of active instances all live there. (The old Fleet Resource Config page now redirects to that tab.)
• Tasks in InProgress with the same elapsed time for >10 minutes → a single task may be in a long retry loop. Open it; the log shows what's happening.

A task shows "TurnLimitReached"

The coder used up its turn budget — the per-run cap on how many steps it may take — before declaring the work done. It's a stop, not a crash: whatever the coder committed is on its branch, but the task didn't finish. On the mission page, the task's row offers the ↻ (Retry task) button — click it to queue the task again for a fresh coder. If the same task keeps hitting the cap, the work is probably too big for one task; ask for a smaller slice in the conversation.

A task sits in "UsageLimitWait"

The provider's usage quota is exhausted for the current window. This is not an error: the task is paused, not failed, and it resumes on its own when the quota window resets. There's nothing to click and no support case to file — leave it alone and check back later.

A mission shows "Paused"

Someone clicked Pause on the mission page — it holds new work without losing anything. Open the mission: in the action row at the bottom, Resume replaces Pause while the mission is paused. Click it and the mission returns to InProgress.

"The action was refused" (invalid transition)

You asked a mission for a state change it doesn't allow from where it stands — the request is refused rather than silently dropped. The error message names the mission's current state and lists the transitions it will accept. Read that list before retrying: the same click will bounce again, and one of the listed actions is the legal next move.

Review screen is empty even though a task completed

• The review may still be syncing — refresh after 30 seconds.
• If still empty, the coder may have produced a no-op (rare but possible). The task page shows what the coder actually changed; if "nothing", that's the cause.

Push to origin failed after approval

• Check the attention request on the review — it usually names the error (auth, network, ref conflict).
• Re-add credentials in the Projects page if the token expired.
• There is no push-retry button — pushes retry automatically, and a task whose push keeps failing is re-queued on its own. Fix the credentials, then give it a few minutes; if the task ends up Failed anyway, the ↻ (Retry task) button on its row is the manual lever.

Workflows

Symptoms specific to missions that run a workflow. For the full picture, see Runs, acceptance & changements.

Cost surprises

• Open the Costs dashboard, filter to the period in question.
• The per-mission view shows which mission and which call site spent the money.
• If a single task ate a lot, open it and look at the model — sometimes a fallback chain bumped you onto a pricier model without you noticing.

I can't sign in

• Confirm the URL: https://<your-slug>.ards.etiakorp.com/.
• Try a private browser window (rules out a stale cookie).
• If you set up a new password recently, give DNS / TLS another minute.
• Still stuck after 5 minutes → file a support case (email us if you literally can't open the case page).

Something else looks wrong / this doc is incomplete

This doc is v1. Gaps and inaccuracies are likely. Please tell us — see Getting help. Useful reports: what you were doing, what you saw, a rough timestamp, and a screenshot if it's visual.