Analyze Mattermost attachments with OpenAI vision for AI agents

Overview

Subworkflow for analyzing Mattermost attachments.

Per-attachment behavior

• Image (mime starts with image/, size ≤ MAX_IMAGE_SIZE_BYTES): downloaded and analyzed by OpenAI vision model.
• Text small (text-like by mime/extension, size ≤ MAX_TEXT_SIZE_BYTES): downloaded as plain text and inlined into the context block (with line count check ≤ MAX_TEXT_LINES).
• Text too large (text-like, exceeds byte or line limit): replaced with a "too big" marker so the agent asks the user for the relevant excerpt.
• Other (binaries, oversized images, unknown): replaced with a generic "not analyzed" marker.

Setup checklist

1. Set MATTERMOST_BASE_URL in the Config node.
2. Create an HTTP Header Auth credential called "Mattermost PAT" with:

• Name: Authorization
• Value: Bearer <your_personal_access_token>

1. Attach this credential to all three HTTP Request nodes (Get file info, Download file, Download text file).
2. Attach your OpenAI credential to the "OpenAI: Analyze Image" node.
3. Tune in Config: VISION_MODEL, MAX_IMAGE_SIZE_BYTES, MAX_TEXT_SIZE_BYTES, MAX_TEXT_LINES.

Notes

• The IF guard short-circuits when file_ids is empty so the parent workflow can call this subworkflow unconditionally.
• Classification by mime + extension catches .log files served as application/octet-stream.
• Final output is always a single item with the same shape — safe to consume in the parent.