Improved compatibility with corporate firewalls, camera AI model import, custom model parsers.

New

Camera Assist : Objects from AI-enabled cameras' ONVIF metadata are now added to Lumeo meta. This is disabled by default, but can be enabled through Video Source node properties. See Camera AI Models for details on camera compatibility.
Improved corporate firewall compatibility
- Add proxy server support for corporate networks via HTTP_PROXY/HTTPS_PROXY/NO_PROXY environment variables. Use lumeo_gateway_manager set-proxy to enable or lumeo_gateway_manager clear-proxy to disable.
- pkgs.lumeo.com is now used for docker pulls (on the host) and pip installs (inside containers).
Clip & snapshot related enhancements
- Added thumbnails for locally stored clips and snapshots.
- Video source node now inserts current clock time in video source node’s metadata (UTC) (similar to frame_number). Clip nodes use this timestamp in filenames, so if a similar trigger fired in another part of the pipeline, it would be possible to correlate between the frame timestamp and the clip file by UTC timestamp.
- In Save Clip and Save Snapshot, metadata fields are surfaced for S3 prefix templating (e.g. {{meta.nodes…}}) to aid event correlation.
Custom model parsers: Support injecting optional Python pre-processing and post-processing nodes around each AI model when the model parameters include the respective 'pre-processing' or 'post-processing' snippets.
Improved host file system management
- Data directory maintenance system with automatic cleanup and retention policies
Installer updates
- Add option to enable unattended upgrades in Lumeo Gateway Manager
- Prompt for Lumeo Web Gateway Manager installation during lumeod install (defaults to 'no')

Fixed

Improved GPU driver error detection by adding nvidia-smi check to catch driver issues that NVML API calls might miss.
Increase ONVIF request timeout from 5 to 30 seconds to support VMS with a large number of cameras.