Power Automate

Power Automate integration in MeinGPT

Two Core No-Code Interaction Modes

  1. Power Automate -> MeinGPT: Flows call assistants and workflows in MeinGPT. This works today via the generic Workflow API or Assistants API — call the API from an HTTP action with your MeinGPT API key (see Authentication).
  2. MeinGPT -> Power Automate: use Power Automate actions as assistant-usable tools.

Note

Interaction mode 2 (Power Automate actions as assistant tools) is coming soon. Mode 1 is already live — use the API docs linked above with any Power Automate HTTP action.

Running an assistant with a file attachment

To trigger an assistant run via API and pass along a file, your HTTP action in Power Automate calls POST /api/assistants/v1/{assistantId}/run as a multipart/form-data request:

  • messages - the message(s) as a JSON string
  • files - the file(s) as binary data (max. 50MB per file, max. 10 files)

See the Assistants API reference for full details and an example.

MeinGPT makes the attached files available to the assistant in two ways:

  1. Full inline content: for text-based formats (JSON, CSV, plain text, XML, YAML), the entire file content is inlined into the assistant's context - regardless of file size.
  2. Document Engine lookup: the assistant is also told the file is stored in Document Engine and can be fetched with a tool. This fallback only works if the assistant has Code Sandbox enabled as a tool.

Note

For assistants that process large file attachments through Power Automate, enable Code Sandbox on the assistant even if it isn't asked to run code. This gives the model a working fallback tool in case the inlined content ever isn't enough.

Non-text formats (images, PDFs, Office documents) follow the same native-file handling described in Code Sandbox - attach Code Sandbox to the assistant if the automation needs to process these formats.

Running a workflow with a file attachment

To trigger a workflow run via API and pass along a file, your HTTP action in Power Automate calls POST /workflows/v1/{workflowId}/run as a multipart/form-data request:

  • input - the workflow's input variables as a JSON string
  • files - the file(s) as binary data (max 50 MB per file)

See the Workflow API reference for the exact per-file size and file-count limits currently enforced, and for how document outputs (documentUrl) are returned.

Note

Enabling or disabling additional document uploads in the workflow editor only affects the chat UI - API calls can always attach files regardless of that setting.

Returning downloadable files from assistants

When Power Automate calls a MeinGPT assistant through the API, the assistant can return downloadable files if it uses Code Sandbox file tools.

To use file downloads in a Power Automate flow:

  1. Enable Code Sandbox on the assistant.
  2. In the prompt or the action input, explicitly ask the assistant to create a downloadable file, for example: "Analyze the attached Excel file and return a downloadable CSV summary."
  3. Call POST /api/assistants/v1/{assistantId}/run from Power Automate, optionally with uploaded files.
  4. Read the response field files. If the assistant exported a sandbox file, the first file is available at files[0].downloadUrl.
  5. Call files[0].downloadUrl from a further HTTP action (GET), or pass the URL into a downstream step that accepts a file URL.

If the assistant only answers with text, files is returned as an empty array. MeinGPT does not create a document from the final text automatically; the assistant must export a file through Code Sandbox.

Next steps

Was this page helpful?