> ## Documentation Index > Fetch the complete documentation index at: https://getskydive.com/docs/llms.txt > Use this file to discover all available pages before exploring further. # Troubleshooting errors > What each error message means and what to do about it When a run fails, the message you see is trying to tell you something specific. This page decodes every error message Skydive shows, in one place. If the message names a fix, follow it. Everything here works the same on every channel: the wording is identical in web chat, Slack, email, and text. ## What to try first One retry is almost always safe. If a retry produces the same error, use the table below to find the exact message and follow its fix. Errors that say "try again in a moment" usually clear on their own within minutes. ## Run errors These appear in the conversation where the run failed, and in run history. | Message | What it means | What to do | | - | - | - | | **This conversation is too long for the model. Start a new thread to continue.** | The conversation has grown past what the model can hold at once. Nothing is broken. | Start a new thread. The agent's memory carries over, so it keeps the context that matters. | | **Your agent hit its usage limit and stopped. It resumes automatically in a few minutes; you can also retry now.** | The agent hit its rate limit. This is temporary, not a billing problem. | Wait a few minutes and it picks back up on its own, or retry now if the work is urgent. | | **Your workspace's model provider key doesn't have enough credits for this request. Add credits, choose another model, or contact a workspace admin.** | The workspace's own connected provider (a key your workspace added) ran out of credits. The failure came from your key, not Skydive's. | Add credits at the provider, switch the agent to a different model, or ask a workspace admin to top up the key. | | **Your Claude subscription is refusing requests because it can't verify that model training is off. In Claude, open Settings > Privacy and turn off 'Help improve our AI models', then retry.** | The Claude account your workspace connected is refusing requests until training consent is verified off. The fix is in Claude, not Skydive. | In Claude: **Settings > Privacy**, turn off **Help improve our AI models**, then retry in Skydive. | | **Your request was blocked by the AI provider's content filter. Try rephrasing your request and sending it again.** | The model provider's safety filter rejected the request. | Rephrase and resend. Rephrasing usually clears it. | | **The model is temporarily unavailable. Please try again in a moment.** | The model provider is overloaded or unreachable right now. | Try again in a moment. This clears on its own. | | **The model stopped responding before this request finished. Please try again.** | A generic provider failure: the model cut out mid-run without a more specific reason. | Retry once. If it repeats on the same request, the provider connection may be the problem; try a different model or contact support. | | **The service that runs the agent's computer is temporarily unavailable. Please try again.** | The agent's sandbox couldn't be started at all. Nothing was lost. | Retry in a few minutes. | ## Plan and compute errors | Message | What it means | What to do | | - | - | - | | **Your connected plan can't take this request right now. Wait for its usage window to reset, connect another account, or switch the model to run on Skydive.** | The BYOK plan the model runs on (your own Claude/OpenAI subscription) hit its usage cap. | Wait for the plan's window to reset, connect another account, or switch the agent to a model that runs on Skydive's credits. | | **The agent's computer isn't responding. Nothing was lost.** | The agent's sandbox crashed or was killed mid-run. Completed work up to the failure point is saved. | Retry. If the same task keeps dying this way, the task may be too big for the agent's compute tier; ask for it in smaller steps or size the agent up. | ## Text message failures (iMessage/SMS) On text, failure notices are shorter and written in the agent's voice. They decode to the same errors as above: | Notice | It means | | - | - | | "this conversation has gotten too long for the model to keep replying in it" | Conversation too long. Start a new thread on web. | | "you've hit the rate limit for now, i'll pick back up automatically in a few minutes, or you can retry now" | Usage limit hit. It self-resumes. | | "your workspace's connected ai provider ran out of credits, so i can't take this right now" | Your workspace's provider key needs credits. | | "the model's content filter blocked this one, try rephrasing your message and sending it again" | Content filter. Rephrase. | | "the model is overloaded with traffic right now, try again in a bit" | Provider overload. Wait and retry. | | "sorry, skydive is experiencing some strain so responses might be slow, i'll pick it up soon" | An infrastructure problem on our side (the agent's computer didn't start, died, or timed out). Nothing you did. No action needed; it picks back up. | | "something interrupted my reply, try again in a bit" | An unclassified failure. Retry. | > [!NOTE] > > In a group text, failure notices are suppressed entirely. If a run fails in a group thread, you'll see the agent simply stop without a notice. ## Email domain errors These appear when connecting or verifying a custom email domain. Full setup steps are in [Email](https://getskydive.com/docs/channels/email.md#custom-domains). | Message | What it means | What to do | | - | - | - | | **This domain is already registered** | Either another workspace owns the domain, a deleted workspace's registration is still held (support can reclaim it), or your domain's MX points at another mail provider (like Google Workspace) and needs the coexistence flow. | If your domain runs Google Workspace, follow the [coexistence steps](https://getskydive.com/docs/channels/email.md#google-workspace-coexistence). Otherwise contact support. | | **Taken** (for an address you've never used) | The address was claimed by another workspace at some point. Addresses are permanent and never reissued. | Pick a different address. | | **Domain limit exceeded** | Your workspace hit its custom-domain cap. | Retrying won't help; contact support to raise the cap. | | **Provider error, try again** (when adding a domain) | A retry usually won't fix this: the common causes are a conflicting provider on your domain, an orphaned registration, or a quota cap. | Check your domain's MX records first, then contact support with the error text and your domain name. | | **Verification stuck on "Verifying" with all records showing valid** | A rare authorization issue on our side. | Contact support with your domain name. | ## What to send when contacting support For anything the table doesn't answer, email [support@skydive.com](mailto:support@skydive.com) with: * The exact error text (copy it verbatim from the conversation or run history) * Which channel it happened on (web, Slack, email, text) * What the agent was doing when it failed > [card] [Run history](https://getskydive.com/docs/workspaces/feed.md) > > Revisit past runs and their errors in the feed. > [card] [Usage and billing](https://getskydive.com/docs/account/usage.md) > > Where credit and usage limits show up. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.