When a step in your thunk fails, Thunk shows you an error with three things:

For the most common failures the remediation note is specific to that error. For the rest it is the general advice — "try running the step again; if the problem persists, see our troubleshooting guide or contact support" — and this page is where that link is pointing. So if the note you got looked generic, find your error below; it will have a specific answer even when the chip did not.

This page explains every error Thunk can raise: what it means, why it usually happens, and what to do next.

Two rules of thumb cover a surprising share of failures:

When one failure happens inside another, Thunk shows you both — the outer step that was running, then the underlying cause, joined by a dash:

Could not download URL for parameter "invoice". - Access denied fetching or updating https://… The Microsoft 365 service account does not have permission to access this SharePoint site or document. Ask the site owner to grant access.

The part after the dash is the one to act on. The chip agrees with it — here you would see Connection/Tool, not User Error, because a permissions problem is the real failure and the download was just where it surfaced. If the first half of a message looks like a dead end, keep reading.

The AI model or the provider serving it failed. Most of these are temporary. The ones that are not are almost always a model-choice problem you can fix in AI Settings (per step, or for the whole thunk).

Why it happens. The AI provider's content filter blocked the request. Something in the step's instructions, or in the data the step was working on, tripped a safety classifier. This is the provider's filter, not Thunk's — and it can fire on innocuous business content (medical, legal, or HR text is a common false positive).

What to do. Rephrase the step's instructions or the input to avoid the content the provider blocks, then run the step again. If a legitimate business document keeps getting filtered, switch the step to a different AI model in AI Settings.

Also appears as: "Your AI model is set to X. This model does not support PDF attachments."

Why it happens. The step tried to send a PDF to a model that cannot read PDFs. Not all models accept PDF attachments.

What to do. Switch this step or the thunk to a model that supports PDF attachments — we recommend a Gemini model — in AI Settings, then run the step again.

Why it happens. The step attached an image to a model that only accepts text.

What to do. Switch this step or the thunk to a model that supports images in AI Settings, then run the step again.

Why it happens. The model caps how many images it will accept in a single request, and the step exceeded that cap.

What to do. Switch to a Gemini model in AI Settings — Gemini models accept far more images per request — then run the step again. Alternatively, split the work across several steps so fewer images go into each request.

Also appears as: "The AI model X request was stopped because the step ran out of time", "timed out after 5 minutes", "request timed out (408)".

Why it happens. The model took too long to respond. This is nearly always the provider being slow or overloaded, not a problem with your thunk.

What to do. Run the step again in a few moments. If it keeps timing out on the same step, the step may be asking for too much in one go — consider giving it less data or splitting it into smaller steps.

Why it happens. Too many requests were sent to the AI model in a short window — either by your workflows or by other traffic sharing the same model deployment.

What to do. Wait a moment and run the step again. If your thunks are rate-limited routinely, ask your administrator about the model's provisioned throughput, or move heavy thunks to a different model.

Why it happens. The model your thunk asked for has no working deployment behind it right now.

What to do. Switch to a different AI model in AI Settings, or ask your administrator to check model availability.

Why it happens. The subscription or API key behind this model has used up its quota.

What to do. Ask your administrator to check the API quota for that model, or switch this thunk to a different model in AI Settings.

Why it happens. The credentials configured for the AI provider are invalid, expired, or lack permission for this model.

What to do. This is an administrator fix, not a thunk fix. Ask your administrator to check the AI provider configuration — the API key and endpoint — in the environment's model settings.

Why it happens. The OAuth token used to reach the AI model has expired and was not refreshed in time.

What to do. Run the step again — a fresh token is normally acquired on the next attempt. If it recurs, ask your administrator to re-authorize the AI provider connection.

Why it happens. The AI provider had a server-side failure. Nothing about your thunk caused it.

What to do. Run the step again in a few moments. If a model returns 5xx errors persistently, switch to a different model in AI Settings and tell your administrator.

Also appears as: "…too long and repetitive, and was cut off mid-tool-call" and "produced tool arguments that are too large for a follow-up request".

Why it happens. The model tried to produce more output than it is allowed to in one response — typically because it was asked to generate or copy a very large amount of data in a single action, or because it got stuck repeating itself. The response was truncated part-way through an action, so the step cannot safely continue.

What to do. Have the step work on less data at a time — narrow the input, or break the work into several smaller steps. If the step is writing a large amount of text into a single field, consider writing it to a file instead. If the model appears to have got stuck in a loop, running the step again often clears it.

Something between Thunk and the service it was calling blocked or dropped the connection.

Why it happens. A network firewall — usually on your organization's side — blocked Thunk's request to the AI model.

What to do. Ask your network administrator to allow access to the AI model endpoint, then run the step again.

These indicate a defect or a limitation inside Thunk. There is generally nothing to fix in your thunk or your input.

Why it happens. Thunk sent a malformed request to the model. This is usually a transient encoding problem.

What to do. Run the step again. If it keeps happening on the same step, contact support — this is a Thunk defect.

Why it happens. The step, or the row it was working on, was deleted while it was still running.

What to do. No action is needed unless you expected the item to still exist. If you did not delete it and it keeps disappearing, contact support.

Why it happens. Thunk could not produce or fetch the internal link to a file the step needed.

What to do. Run the step again. If it keeps failing on the same file, contact support with the thunk and step name.

Why it happens. The thunk is using a tool that has been retired from the platform.

What to do. Remove or replace the step that uses the retired tool. Contact support if you need an alternative approach for the same work.

Why it happens. A document template is being filled with data whose field name collides with a name the template engine reserves for its own use.

What to do. Rename the field — in both the template and the data feeding it — so it no longer collides.

Something about how this thunk is configured is wrong. These are for the thunk author to fix, and they will keep happening until the configuration changes — retrying will not help.

Why it happens. The thunk (or a specific step) asks for an AI model that this environment does not offer, and there is no equal-or-better model from the same provider to substitute. The error lists the models that are available.

What to do. Pick one of the listed models in AI Settings, or ask your administrator to add the model you want to the environment.

Why it happens. The model deployment configured for this environment does not exist on the provider side — usually a typo or a deployment that was deleted.

What to do. Ask your administrator to verify the model configuration in AI Settings.

Why it happens. Web search is only possible with a Gemini model — search grounding is a Gemini-only capability. This environment has no Gemini model configured.

What to do. Ask your administrator to add a Gemini model to your environment. Until then, the web search tool cannot run.

Why it happens. A document-template tool was created without pointing it at a template.

What to do. Open the tool's settings and set a Google Doc or Word template URL.

Why it happens. Someone edited the template document — added, removed, or renamed a {{placeholder}} — after the tool was built. The tool's saved schema has drifted from the live template.

What to do. The tool's author must open the tool builder and re-fetch the template variables, so the tool picks up the current set of placeholders.

Why it happens. The template contains malformed {{...}} syntax.

What to do. Fix the syntax error reported in the message, then try again.

Why it happens. A placeholder is written like {{Embassy Address}}. The template engine reads a space as a function call, not as one field name.

What to do. Rename placeholders to use only letters, numbers, underscores, and hyphens — {{EmbassyAddress}} or {{embassy_address}}.

Why it happens. A placeholder uses characters the template engine cannot treat as a field name.

What to do. Use only letters, numbers, underscores, and hyphens. For nested fields, use dots: {{person.name}}.

Why it happens. The template has no {{...}} fields at all, so there is nothing for the tool to fill in.

What to do. Add placeholders to the template — fields like {{customerName}}, or block sections like {{#each items}}...{{/each}}.

Why it happens. The template file itself could not be opened to read its text — it is corrupt, truncated, or is not really a .docx.

What to do. Open the template in Word and re-save it as .docx, then point the tool at the re-saved file. Note this is filed as a Thunk Definition error, not a User Error, because the file at fault is the thunk's configured template — not something supplied at run time.

Why it happens. Same root cause as the "names with spaces" error above, but caught at render time: a placeholder name collides with a reserved helper name, or contains a space.

What to do. Rename the offending field in both the template and the data. The error message lists the reserved names to avoid.

Why it happens. Either the URL points at something that is not a document, or the connection module needed to read it (Google Drive, or Microsoft 365) is not enabled for that tool.

What to do. Check the URL is a Google Doc or a .docx. Then enable Google Drive or Microsoft 365 in the tool's Module config and try again.

Why it happens. The thunk's conversation is configured to deliver over Teams, which has been retired as a channel.

What to do. Change the conversation channel — to email, Slack, or Google Chat — in the thunk's conversation configuration.

Why it happens. The thunk's conversation is configured to deliver over the GiGi channel, which is not sent by the thunk itself — GiGi messages are delivered through the GiGi chat integration instead.

What to do. If you meant the thunk to send the message directly, change the conversation channel to email, Slack, or Google Chat. If you are using GiGi, the message is delivered through that integration and no action is needed here.

Why it happens. The conversation is configured to deliver to Google Chat, but no space was chosen.

What to do. Set the Google Chat space address in the thunk's conversation configuration.

Why it happens. A tool has validation rules ("checkers") attached, and the agent could not produce a tool call that satisfies them — after several attempts. Usually the rules and the step's instructions are pulling in different directions, or a required input is missing.

What to do. The error lists which checks failed. Review the step's instructions and the tool's rules and make them consistent — often the step is missing a piece of information the check requires.

Why it happens. A filter or condition in the thunk uses an operator the evaluator does not know.

What to do. Re-open the filter in the workflow editor and rebuild it using the operators offered in the picker.

Why it happens. A step being added would break the chain of statuses that connects the workflow's steps in order.

What to do. The message names the specific problem. Review the statuses on the surrounding steps so each step's finishing status matches the next one's starting status.

An external system, credential, or document was the problem. Retrying rarely helps until the underlying access is fixed.

Why it happens. The thunk was handed a Google Drive or OneDrive/SharePoint link, but there is no matching connection available for it to use. Without a connection, Thunk cannot see anything a signed-out visitor cannot see.

What to do. Three things to check, in order:

Why it happens. The Microsoft 365 service account Thunk uses has not been granted access to that SharePoint site or document.

What to do. Ask the SharePoint site owner to grant the service account access to the site or document. This is a permissions change on the Microsoft side — not something you can fix in Thunk.

Why it happens. The link points at a SharePoint site belonging to a different organization than the one connected to Thunk.

What to do. Check that the template or source URL belongs to the correct organization. A link copied from a personal or partner tenant will not work.

Why it happens. The URL does not resolve to a document Thunk can open — it may be wrong, the file may have moved or been deleted, or the connected account may not have access.

What to do. Verify the link opens for you. For Word, it must be a valid OneDrive or SharePoint URL for a .docx file. For Google Docs, a valid Google Docs URL.

Why it happens. The document exists, but the Google service account Thunk uses cannot open it.

What to do. Share the document with the service account, or move it into a shared drive the service account can already reach.

Why it happens. Every file the step tried to upload to OneDrive or Google Drive was rejected individually. The message lists the per-file reasons — most often the connected account lacks write permission on the destination folder.

What to do. Read the per-file reasons in the message, and check that the connected account can write to the destination folder.

If only some files failed, the step does not stop — it carries on with a partial result. It will usually still finish as an error, but not always. Whenever an upload step reports a partial failure, check which files actually landed before treating the work as done.

Why it happens. The email provider has suppressed that address because of an earlier delivery problem: a hard bounce (the address does not exist), a spam complaint, or an unsubscribe. Once an address is suppressed, further sends to it are blocked automatically.

What to do. First verify the address is spelled correctly — a typo that hard-bounced once will stay suppressed. If the address is correct and should be receiving mail, contact Thunk support and ask them to investigate the suppression; it cannot be lifted from inside the product.

Why it happens. The email provider rejected the send for some other reason, or was unavailable.

What to do. Run the step again. If it persists, contact support.

Something supplied to the thunk — a file, a URL, an address — is unusable as given. The fix is in the input.

You may see any of:

Why it happens. The files are larger, or more numerous, than a single AI request or upload can carry.

What to do. Reduce the size or number of files. Practical options: attach fewer files per step and process them over several steps; compress images before attaching; or split a large document and work through it in sections.

Why it happens. The file was sent as a PDF but could not be parsed — it is corrupt, truncated, or is not actually a PDF despite its extension.

What to do. Open the file yourself to confirm it is a valid PDF. Re-export or re-download it and attach it again.

Why it happens. The .docx could not be parsed — it is damaged, truncated, or in a format that is not really a Word document (an old .doc renamed to .docx, for example).

What to do. Open the document in Word and re-save it as .docx, then try again.

Why it happens. The workbook is too large for Thunk to load into memory. The message tells you the size it found and the limit it enforces.

The limit is on the whole file, and it is checked before Thunk opens the file at all. So asking for less of the workbook — fewer rows, or one sheet — cannot get past it. The file itself has to get smaller.

What to do.

Why it happens. The sheet is wider than Thunk can read in one pass.

What to do. Remove columns the workflow does not need, or export a narrower view of the data and point the step at that.

Why it happens. The file has more rows than can be read in one pass.

What to do. Ask the step to read only the first N rows if partial data is enough. Otherwise, split the CSV into smaller files and process them across several steps.

Why it happens. A file was added to a document collection in a format Thunk cannot index and search.

What to do. Convert the file to a supported format — PDF, Word, or plain text are the safe choices — and add it again.

Why it happens. Something that is not a spreadsheet Thunk can open was added to a Spreadsheets collection.

What to do. Only Google Sheets and SharePoint/OneDrive spreadsheet URLs, or uploaded .xlsx, .xls, and .csv files, can go into a Spreadsheets collection. Convert or re-link the item accordingly.

Why it happens. The SharePoint URL format used is one Thunk cannot resolve unambiguously to a single file.

What to do. Open the file in SharePoint and use its Share link, rather than copying the URL from the browser address bar.

Why it happens. The folder URL given to a step is either not a recognisable OneDrive/SharePoint folder URL, or it points at a file.

What to do. Provide the URL of the folder you want to write into, not a file inside it.

Why it happens. The URL could not be fetched, or the server returned a response Thunk cannot classify — often a login page, an expired link, or a page that requires a connection Thunk does not have.

What to do. Confirm the URL opens in a private browser window without signing in. If it needs credentials, add the appropriate connection and enable it for the thunk and the step.

Why it happens. A URL passed into a tool could not be fetched.

What to do. Check the link is correct and publicly reachable, or that the connection needed to read it is enabled for this thunk and step.

Read past the dash. For Drive and SharePoint links this error usually carries an underlying cause after the first sentence — most often a permissions failure, which is chipped Connection/Tool. If so the link is fine and the access to it is not; go to the Connection/Tool errors section.

Why it happens. The thunk's conversation configuration contains an address that is empty, or that is missing an @.

What to do. Open the thunk's conversation configuration and correct the email recipients so every address is a valid email.

Why it happens. A file the email was supposed to carry could not be downloaded or converted for inclusion.

What to do. Check that the attachment still exists and is reachable. If it lives in Drive or SharePoint, confirm the connection is enabled and the connected account has access.

Read past the dash. As above — if the underlying cause is a permissions failure (chipped Connection/Tool), the attachment exists and Thunk cannot reach it. That is a permissions fix, not a file fix.

If none of the above resolves it — or the error is chipped Other, meaning Thunk did not recognize it: