Skip to main content

Troubleshoot a failed task

Yekar.AI names its failures, so almost every problem tells you which knob to turn. Start with what you see in the interface. Flow errors appear on the failed step in the session’s Run tab; Agent errors appear in the conversation and Timeline. Mutation failures also appear as an error notification.

The integration is unavailable​

You see INTEGRATION_NOT_VISIBLE. The integration is not available to this organization, or a managed platform integration was removed. Ask an administrator to make or install it for the organization, then retry.

You see PLATFORM_APP_UNAVAILABLE. A built-in capability is unavailable for your workspace. Contact Yekar.AI for help enabling it.

A connected account is missing​

You see SERVICE_CONNECTION_MISSING. The agent or flow runs as its domain's team account and that domain has no usable connection. Its Integrations (agent) or Connections (flow) drawer says which domain it expects and links straight to the place to fix it; otherwise a domain owner opens Integrations → Catalog, chooses the integration, picks that domain under Acting as, and connects it.

A banner says the agent runs as you and you haven't signed in. The agent runs as the person using it, and you have no sign-in for that integration. Follow the link in the banner, or open Integrations, choose the integration, pick My account under Acting as, and sign in. Nothing stops you sending the first message: on Fall back to the team account the turn runs on the team account and says so, and otherwise that integration's tools are simply absent from the turn.

A banner says a domain owner has turned off personal sign-in. The agent runs as the person using it. Personal sign-in is on by default, so this means an owner switched it off for that integration in this domain - and no fallback reaches past a door an owner has shut. A domain owner opens the integration's page for the domain and turns Members may sign in with their own account back on.

A scheduled run was skipped with person_unavailable. The agent runs as the person who created the trigger, and they are no longer an active member, haven't signed in, or an owner has turned personal sign-in off for that integration in the domain. Re-create the trigger as someone who is, or set the binding to the team account.

You see SANDBOX_CREDENTIAL_MISSING. A Flow test selected a named credential that was deleted before the step ran. Open Test run or Test this step, choose an existing credential, and retry.

An API caller credential cannot be used​

You see CALLER_CREDENTIAL_REQUIRED. The execution requires an API caller token, or stored-account fallback was forbidden. Fire it again with the required token. A support session without a token cannot use the organization’s stored credential.

You see CALLER_CREDENTIAL_EXPIRED. The supplied caller token expired. Fire the agent or flow again with a fresh token; an expired supplied token never falls back to a stored account.

You see NO_CREDENTIAL_AVAILABLE. No caller token was supplied, fallback was permitted, and the domain has no stored credential. Supply a caller token or connect a Team account for the domain.

You see CALLER_GRANT_CORRUPT. The saved grant for this execution cannot be read safely. Do not retry the same execution with fallback. Start a new call with fresh caller credentials; if it repeats, ask an administrator to investigate the execution record.

The model cannot start​

You see LLM_NOT_CONFIGURED. The organization has no usable key for the Agent’s selected model provider - either none was added, or the provider is set to Off. An organization administrator should open Settings → AI settings → AI providers, confirm the provider is set to Own key, add the provider key, and retry.

You see LLM_MANAGED_UNAVAILABLE. The provider is set to Managed, but Managed access is currently unavailable. Contact Yekar.AI, then retry.

You see LLM_MANAGED_RATE_CARD_UNAVAILABLE. The selected model has no effective managed rate card. Choose another available model or contact Yekar.AI.

The conversation says an image attachment was omitted. This is not an error - the session's model does not accept image input, so the image was left out of the turn and the message says so, naming the model. The text of the message still went through. To work with images, pin the Agent to a vision-capable model under Model & quality, or ask an administrator to enable one - see Image input.

A Procedure knowledge reference is unavailable​

A file or article is missing from the / picker. Files must be attached to this Agent and finish extraction. Articles must be active, published, finished processing, and included in the Agent's Knowledge selection. Under Organization, articles from another domain need Organization visibility. Save any selection changes, then close the picker and type / again.

Verification or a turn says a referenced source can no longer be read. Restore the source and its access, finish processing or publishing it, or remove the reference from Procedure and publish the change. Referenced knowledge requires Standard execution mode.

A turn says referenced knowledge exceeds the context limit. Use smaller sources, remove references that do not need to be included on every turn, or increase the context limit. Removing a Procedure reference leaves an attached file or selected article available for normal retrieval. See Procedure references.

A budget or a limit stopped it​

The conversation says I can't respond right now and names a budget. The organization's or the domain's monthly token budget is spent (LLM_BUDGET_EXHAUSTED), or it could not be checked (LLM_BUDGET_UNAVAILABLE). An organization admin raises or clears the budget under Settings → AI settings → Defaults, or a domain owner on the domain's page; a change bites within about a minute. Nothing mid-flight was cancelled. See Cost & budgets.

You see MANAGED_CAP_EXCEEDED or MANAGED_BALANCE_EXHAUSTED. Managed spend passed the monthly cost cap, or the prepaid balance cannot cover the next call. Raise the cap, or add credit.

You see AUTOMATION_AT_CAPACITY. The agent or flow is at its published Max concurrent runs. Wait for a run to finish, or raise the limit and publish - a draft value does not apply.

You see ORG_RUN_RATE_LIMITED. Too many runs were started in one minute for the organization's plan. Retry after the Retry-After seconds. Subagents started by an agent do not count against this.

A called agent was refused with AUTOMATION_CALL_TOO_DEEP, AUTOMATION_CALL_TOO_MANY_CHILDREN, AUTOMATION_CALL_TREE_TOO_LARGE or AUTOMATION_CALL_CYCLE. A call tree hit one of its limits. See Calling other agents.

An approval is stuck​

The row says Credentials required. The credential the gated call runs on became unavailable, or the person it was routed to left the domain. Restore the credential - Update credentials on the row - and the block clears on the next sweep. See How approvals are layered.

The approval expired. Nobody decided within the window, and the work failed rather than running unapproved. Run it again; raise the timeout in the Agent's Approvals or the flow step if the window is too short for your approvers.

A schedule did not run, and its trigger shows a count on Approvals. Occurrences the platform could not fire on time are parked as a catch-up batch on the trigger's Approvals tab until someone approves or discards them.

Still stuck?​

If the code shown in the Run, Timeline, or notification is not listed here, keep the exact code and message, note the agent or flow and the session, and give those details to an administrator. Do not replace a missing credential with a more privileged account simply to make the retry pass.