Release notes - 1.4.0
Release date: September 1, 2026
Overview
Version 1.4.0 ships Lucia's first official appliance image for the NVIDIA
Jetson Orin Nano Super 8GB Developer Kit. It installs a native, Docker-free
Lucia system from microSD to NVMe, including local voice models, CUDA runtime
support, Redis, SQLite, the dashboard, and appliance management.
This release also restores deterministic command routing for unresolved entity
names and adds explicit llama.cpp provider support.
Appliance image
- Install from a flashable microSD image onto an NVMe drive of at least
61,203,283,968 bytes. The installer binds erase approval to the selected
drive's stable identity and image digest before writing it. - Run Jetson Linux 36.5.2 from A/B operating-system slots, with separate
partitions for versioned Lucia files and persistent application data. - Complete setup through a client-isolated captive network. The first browser
claims the setup session, selects Wi-Fi, chooses the hostname and recovery
password, and receives the dashboard API key. - Recover from a failed Wi-Fi activation without reprovisioning. The installer
rolls back the NetworkManager checkpoint and restores setup mode. - Use the native Jetson GPU for Lucia's bundled speech models. The image
includes the pinned ARM64 ONNX Runtime, CUDA provider, sherpa-onnx libraries,
and voice assets needed for offline setup. - Open the installed dashboard at
https://HOSTNAME.local:8099. Each appliance
creates its own certificate during setup, so the first browser must accept
the local certificate.
See the appliance release guide for the
asset layout and release details. (#257)
Appliance management and telemetry
- View appliance, storage, service, Wi-Fi, and operating-system status from the
dashboard. - Restart Lucia services or reboot the appliance through the authenticated
AgentHost adapter and root-owned appliance manager. - Configure authenticated OTLP export without exposing telemetry credentials
in appliance status responses. - Run the OpenTelemetry Collector and Redis exporter when remote telemetry is
enabled. Both remain disabled by default. - Discover compatible Lucia and operating-system releases from GitHub.
Installing discovered updates remains locked in this release.
Release integrity
- Stable releases publish separate installer, Lucia, and operating-system
channels. - Large images are split into GitHub Release parts below the 2 GB asset limit.
lucia-appliance-manifest.jsonrecords compatibility, sizes, SHA-256 hashes,
ordered parts, and download URLs.- GitHub build-provenance attestations cover every release asset. The workflow
uploads the manifest last so an incomplete release cannot be discovered by
an appliance.
Command routing
- Retry unresolved locations as entity names through the configured embedding
provider instead of dropping out of the deterministic command path. - Recognize direct climate commands such as "Set the office to 73" with enough
confidence to use the fast path. - Add an explicit llama.cpp provider while preserving existing OpenAI-compatible
endpoint behavior. (#256)
Read before installing
- The image supports the Jetson Orin Nano Super Developer Kit with the 8GB
P3767-0005 module. Other Jetson boards and carrier configurations are not
supported by this release. - Installation erases the selected NVMe drive after explicit confirmation.
Keep the appliance physically controlled while its open setup network is
active. - The dashboard can discover updates but cannot install them yet. Attestation
verification, application rollback, and NVIDIA A/B OTA apply are still in
progress. - Automatic QSPI compatibility reporting, full occupied-drive layout
reporting, physical power-cut recovery, secure boot, and disk encryption are
not included in v1.4.0.
Breaking changes
None. Appliance mode defaults to Off, so Docker and other existing
deployments keep their current routes and behavior.