Limits & large payloads

superglue tools can fetch, parse, and forward large files. This page lists the limits that apply, what happens when data crosses them, and how long large payloads take to process.

HTTP responses (downloads)

A tool step can download responses up to 4 GB. What happens next depends on the size and content:

Response	Behavior
JSON, XML, or CSV up to 4 GB	Parsed into structured data your next steps can transform and query
Other text formats up to 500 MB	Parsed into structured data your next steps can transform and query
Anything larger, or binary	Kept as a file. Later steps can forward the raw bytes or process them in code
Larger than 4 GB	The step fails with a size error

JSON, XML, and CSV are read incrementally, so they parse at sizes no single block of text could hold. Other text formats (YAML, HTML) are read in one piece and stop at 500 MB.

Archives and documents (ZIP, Excel, PDF, Word, PowerPoint) are opened and parsed at any response size. Compressed content stops expanding at 4 GB.

Parsing is fast at any size: a 100 MB response parses in about one second, whether XML, JSON, or CSV. Transfer time over the network is what dominates large downloads.

A payload can still exhaust the memory available to a run before it finishes parsing. When that happens the step keeps the file, reports the reason, and the run continues. The bytes stay available through .raw.

HTTP uploads (requests)

A tool step can send a stored file as a request body with a file reference such as file::report.raw. Uploads declare their exact file size, so they work with services that require it, such as S3.

• Maximum upload size: 4 GB per request.
• Larger files fail immediately with a clear size error. No partial upload starts.

The typical pattern for moving big files: one step downloads a file from system A, and a later step in the same run sends it to system B via file::…​.raw.

Moving data out of a run

Files and structured results that leave a run are size-limited:

• Tool results (structured data): up to 500 MB.
• Files returned to the platform or the agent: up to 375 MB per file, sharing a 500 MB budget with the result. Files are encoded for transport, which adds a third to their size, so a single file near 375 MB fills that budget on its own.

Multi-gigabyte files stay inside the run: fetch them and forward them in workflow steps instead of returning them as results.

Direct file uploads

Files uploaded through the API or the chat interface (for parsing and extraction) are limited to 375 MB per file. API request bodies are limited to 500 MB.

Timeouts

Every request in a tool run times out after 60 minutes by default. For HTTP requests this includes the full download, so a large transfer must complete within the timeout.

You can change timeouts in two places:

• Organization admins can set timeouts per protocol under Settings → Timeouts. This requires a plan with the custom timeouts feature.
• A single tool run can pass its own timeout in the run options, which overrides the organization setting for that run.

Error behavior

Every limit fails fast with an explicit error message that names the limit, so runs never hang on oversized data:

HTTP response exceeds the 4 GiB limit
File file::export.raw is 5368709120 bytes. HTTP uploads support files up to 4 GiB.
Decompressed content exceeds the maximum of 4294967296 bytes
Parsing this payload ran out of memory after 9830 MB. Keep the file in the workflow and forward it with .raw instead of parsing it.