Import from connected services
Connectors copy a selected file into Mizura's local dataset store. They do not keep a live link to the remote file; import again to get later changes.
Choose a source
Start in the workspace dataset panel and choose Browse Google Drive, Browse GitHub, or Browse Zenodo. These actions load the selected file into the current session.
| Source | Use it | Access | Limits and notes |
|---|---|---|---|
| Zenodo | Enter a DOI or record URL, select a listed file, then import. | No account; public records only. | Files are capped at 200 MB by default. The selected filename must have a supported dataset extension. |
| GitHub | Paste the repository home URL. Select a file from the repository's latest default-branch commit. | Public repositories need no sign-in. Connect GitHub through device authorization for private repositories. | At most 500 matching files are shown. The import uses the commit selected during listing, so that imported copy is pinned even if the branch moves later. GitHub App configuration may be required. |
| Google Drive | Connect, search/list files, then import a file or Google Sheet. | Google sign-in is required; requested Drive scope is read-only. | Sheets export as CSV. Files are capped at 200 MB by default. OAuth client configuration may be required. |
Imports use HTTPS to the provider's allowed hosts, check the size while downloading, and refuse redirects to other hosts. A bad or unreadable file is rejected and removed rather than added as a dataset.
- For Zenodo, enter a DOI or record URL and select Browse. For GitHub, enter the repository home URL and select Browse. For Drive, sign in if needed and optionally search the file list.
- Select the supported file in the results. GitHub shows the pinned commit above the list. A Google Sheet is imported as CSV.
- After the picker closes, check that the dataset preview shows the expected file name, row count, columns, and first rows. Inspect it before analysis.
If it fails: read the message shown in the picker. “No supported dataset files” means the provider listing had no supported extension or the filter has no matches; clear the filter or check the repository/record link. If GitHub says the list is truncated at 500 files, download the desired file from GitHub and use direct upload. “File exceeds the upload limit” means choose a smaller file or use direct upload when its format and direct-upload limit allow it. A corrupt or unreadable file is not added; re-export or convert it, then try again. For authorization errors, confirm the account can read that file and reconnect. If a connector says setup is needed, the engine operator must configure its OAuth client.
What happens after import
- The selected remote bytes are downloaded by the local Mizura engine.
- Mizura reads the file as a table, stores it in its dataset area, and shows its preview.
- The imported dataset carries a source reference when the connector provides one. Later remote edits do not update this local copy.
Connector imports use a 200 MB default cap, configurable by the engine operator. Direct uploads have separate limits: 10 GB for streamed formats and 200 MB for in-memory formats by default. See dataset size limits.
Connections and credentials
To connect or disconnect, open Settings → Connectors. For GitHub, choose Connect GitHub, enter the displayed device code on GitHub, then return to Mizura; a connected state exposes Disconnect GitHub. For Drive, choose Connect and finish Google sign-in; the status changes to Connected. Zenodo needs no connection.
If a device code expires or sign-in is closed before the status changes, start the connection again. If Google shows Setup needed or a connector cannot be configured, contact the person who installed or operates Mizura; the OAuth client is build/server configuration.
GitHub and Google Drive connections are saved by the local engine for reuse across sessions. Tokens are stored in owner-readable JSON files in the engine's data folder; they are not encrypted, and the local engine does not return them in API responses. Disconnect in Settings when access is no longer needed.
Imported data remains local for analysis, but dataset profiles and other conversation context may be sent to the configured model service. See Security & data handling.