Release Notes

• 1: SPEKTRA Edge Release Notes
• 2: SPEKTRA Edge OS Release Notes

1 - SPEKTRA Edge Release Notes

For changes to the operating system running on your devices, see the SPEKTRA Edge OS release notes.

2026-09-25

• Search: Fixed an issue with search results.
• Search: Equality, IN, and CONTAINS filters in search requests match exact values and are case-sensitive, the same as in list requests. Free-text phrase search is unchanged.

2026-09-23

• SPEKTRA Edge Dashboard: Free-text search and structured filters on list views, with autocomplete-style suggestions where the platform supports them.
• SPEKTRA Edge Dashboard: Monitoring items are grouped in the sidebar; project overview shows a clearer resource summary.
• SPEKTRA Edge Dashboard: Home dashboard can show AI alerting cards and personalized starter layouts based on your permissions.
• SPEKTRA Edge Dashboard: AI alert insight cards summarize recent alerting activity.
• SPEKTRA Edge Dashboard: Access pages group grants by principal and show grant date, role detail, group links, and account state.
• SPEKTRA Edge Dashboard: Device details share a metrics time range across charts, with container log filtering and SSH / USB Guard summaries.
• SPEKTRA Edge Dashboard: Application pods support log Live and History views; the pod-detail Logs tab is renamed Audit for platform activity; distribution detail links to the project’s Devices list filtered by that distribution.
• Agent Dashboard: Conversation search runs across projects you can access; dark theme styling is corrected, and you can choose light, dark, or system theme.
• Agent Dashboard: You can resume or resend after an interrupted chat turn.
• Agent Dashboard: Conversation feedback hides after submit even when your sign-in email uses mixed-case letters.
• Agent Dashboard: Agent-preset MCP connector checkboxes no longer select project-scoped and system-scoped connectors together when they share an id.
• ZTP Dashboard: During camera scan, the modal lists only devices found in the current scan session; scanning a QR code from an image file is more reliable while the camera is still running.
• Search: List filters gain autocomplete-style prefix suggestions.
• Devices: Device identity matching rejects placeholder serials and follows a clearer field priority when choosing which identity field to use.
• AI: Troubleshooting no longer drops silently after a conversation is created.
• AI: Claude adaptive thinking and GPT reasoning summaries are available where the model supports them.

2026-08-26

• AI: Cross-project documents used for troubleshooting are limited to an allowlist.
• AI: Troubleshooting sessions recover more reliably when a previous run did not finish cleanly.
• AI: Built-in AI roles include the permissions needed to use AI features end to end.
• IAM: Email verification is tied to the identity claim selected for sign-in.
• IAM: Keycloak and Zitadel can be used as the identity provider, with a switchable user-management setup.
• IAM: Authentication accepts a configurable clock-skew allowance between clients and the platform.
• IAM: Device agents can use ZTNA provider capabilities they need to operate.
• IAM: Projects, Organizations, Groups, RoleBindings, and ServiceAccounts can be filtered by lifecycle state in search; Projects can also be filtered by parent organization.
• Devices: Devices and ProvisioningPolicies can be filtered by lifecycle state in search.
• Applications: Pods, ConfigMaps, and Distributions can be filtered by lifecycle state in search.
• Secrets: Secrets can be filtered by lifecycle state in search.
• Secrets: Secret encryption continues to work through Vault Transit disruptions, with protection against accidental key deletion.
• Alerting: Related pod-health alerts can be grouped the same way device-connectivity alerts are.
• Search: Platform search now uses the built-in search backend only (the previous Algolia-backed option is removed).

2026-08-12

• SPEKTRA Edge Dashboard: AI chat routes and project menu items appear only when you have the required IAM permissions.
• SPEKTRA Edge Dashboard: You can continue an AI conversation from an alert and resume or resend after an interrupted turn.
• SPEKTRA Edge Dashboard: Device pages follow each droplet’s managed-feature settings more closely.
• SPEKTRA Edge Dashboard: Anomaly-detection forms keep raise/silence durations correctly when you edit them.
• SPEKTRA Edge Dashboard: Devices and pods show health-check status at a glance.
• SPEKTRA Edge Dashboard: Visual theme updates across chat, alerts, headers, and shared controls, including clearer charts in dark mode.

2026-07-08

• AI: Troubleshooting jobs time out instead of running indefinitely.
• AI: Agents no longer escalate automatically when a device still shows a stale disconnected status.
• AI: Retired chat models are blocked so clients cannot keep using a model that has been withdrawn.
• Alerting: Alerts flap less often because silence windows follow raise windows by default; simultaneous device-connectivity alerts can be grouped into one.
• Audit: Audit export is more reliable during shutdown, and dropped audit events are visible as metrics.
• Devices: Device status reports which droplet features are enabled.
• Devices: Legacy devices show connection status correctly.
• Devices: TPM manufacturer certificates are available for device attestation flows.
• Limits: Usage can be queried through usage-record APIs, and allowance policies can be configured.
• Monitoring: Custom metrics keep working correctly after API-server restarts.
• Monitoring: Timeseries ingest and query stay available through fixes for a crash loop and a metadata deadlock.
• API: Responses can be larger by default (up to 64 MB), which matters for clients that read big list or export payloads.
• IAM: Bug-report permissions and a dedicated role are available for submitting platform bug reports.

2026-06-10

• AI: Chat models running on your devices can be used through proxies with an OpenAI-compatible API.
• Monitoring: Metric streams stay connected when rollup connector addresses change.
• IAM: Role bindings create the linked managed account consistently, including for shared-catalog resources.
• SPEKTRA Edge Dashboard: A wizard helps set up agent presets, including template variables in prompts.
• SPEKTRA Edge Dashboard: Home dashboard cards are easier to add and arrange; organization and project lists respect your permissions.
• SPEKTRA Edge Dashboard: Prompt templates keep their structure when variable values are filled in; the first chat message is handed off reliably when opening a conversation.
• SPEKTRA Edge Dashboard: Sign-in return URLs stay on the same site for safer redirects.
• Agent Dashboard: Region defaults follow the selected service; device and organization/project region handling is clearer.

2026-05-06

• IAM: Standard roles such as viewer and admin-operator can use AI features.
• Limits: A project can apply one usage policy that covers all AI models.
• Audit: Audit trails capture more of the actions that matter for review.
• SPEKTRA Edge Dashboard: Volume lists in workload forms edit correctly.

2026-04-15

• AI: Token usage can be limited with quotas.
• Limits: Connections keep themselves alive by default; API clients can set unary request timeouts.
• SPEKTRA Edge Dashboard: Alert notification masks behave correctly when editing alerting settings.

2026-04-01

• Devices: Devices and pods are searchable.
• AI: Alerting can call AI and request structured response formats; conversations are fully searchable.
• IAM: Auth0 database users can reset passwords through the platform; runbook-editor and runbook-viewer roles are available.
• Audit: Tool executions are recorded in the audit log.
• SPEKTRA Edge Dashboard: Device upload and loading are more reliable; SSH sessions to devices stay alive; AI chat entry points are clearer.

2026-03-11

• ZTP Dashboard: Device processing statuses and labels are clearer during registration.
• ZTNA: ZTNA is available in production for the first time.
• AI: Applications can call the AI service with a service account.
• Monitoring: CPU utilization metrics appear as expected.
• SPEKTRA Edge Dashboard: Advanced ZTNA pages live under the ZTNA menu; proxy settings validate more clearly.
• SPEKTRA Edge Dashboard: Pod log viewing supports better container selection and streaming.
• SPEKTRA Edge Dashboard: Config maps and secrets can be managed through dashboard actions.
• SPEKTRA Edge Dashboard: Forms surface errors more clearly across common workflows.

2026-02-18

• Devices: Breaking (API) — In Device control-plane interface status, usesProxy is renamed to uses_proxy for JSON/REST clients and field masks. Binary protobuf clients are unaffected because the field number is unchanged.
• Proxies: Connectors work without OAuth; relay connections are more stable and recover cleanly from failures.
• Devices: Allowed clock difference between devices and the platform is configurable.
• AI: Conversation titles with non-ASCII text no longer get cut off incorrectly.
• Monitoring: CPU utilization metrics are available.
• SPEKTRA Edge Dashboard: Editing pod volumes updates the right fields.

2026-01-28

• Devices: Devices may be up to about 10 minutes off from platform time and still authenticate.
• Devices: Spurious metrics no longer appear when devices connect; unmanaged devices are handled as current devices.
• IAM: Projects can turn platform features on or off through feature configuration.
• AI: Conversations can be filtered by template; streaming health checks are available for diagnostics.
• SPEKTRA Edge Dashboard: Project feature configuration for AI is available in the UI.

2026-01-26

• Devices: Devices may be up to about 10 minutes off from platform time and still authenticate.

2026-01-14

• AI: MCP tools can authenticate with OAuth.
• IAM: Admin membership is removed when the last admin role binding goes away; agent assignment stays stable under load.
• SPEKTRA Edge Dashboard: Chat input handles the Enter key more predictably.
• SPEKTRA Edge Dashboard: The permissions table handles organizations correctly; more device statuses have clear labels.

2025-12-08

• ZTP Dashboard: Device registration shows additional device processing statuses with clearer labels.
• AI: Prompt-template and conversation polish on top of the prior AI release, including clearer resume/edit behavior in day-to-day use.
• Alerting: Adaptive thresholds are easier to tune; conditions can use anomaly detection alone.
• SPEKTRA Edge Dashboard: Private chats, project switching, secret management, and conversation feedback are available.
• SPEKTRA Edge Dashboard: Prompt templates show guidance and variable fields in chat; dark mode is improved.
• Agent Dashboard: Generally available release.

2025-12-07

• AI: Conversations handle ownership and errors more clearly; interrupted tool runs can be resumed.
• AI: System prompts can use templates; Responses API traffic routes across regions.
• AI: Breaking (API) — available_tools on conversations is replaced by available_tools_by_source. Older field-mask paths that still send available_tools are rejected.
• IAM: Listing your own role bindings works correctly across regions.
• Audit: Audit log batches respect configured size limits.
• Agent Dashboard: Beta update with stronger conversation and prompt-template support ahead of GA.

2025-10-24

• AI: Tool-call limits are temporarily relaxed; conversation cost display and message content are corrected.
• AI: Breaking (API) — TokenUsage.input_details / output_details are replaced by input_breakdown / output_breakdown. Clients that read the old detail fields must switch to the new breakdown fields.
• Agent Dashboard: First production deployment (beta).

2025-10-22

• AI: Usage and cost are visible; conversations reuse cached tokens where possible; tool limits and cost details are clearer.
• AI: Each agent can be limited to allowed models.
• Alerting: Notifications are delivered more reliably; alerts that suggest rebooting a pod include the pod identity.

2025-09-24

• SPEKTRA Edge Dashboard: AI chat is available in the dashboard; remediation actions consider device status.
• SPEKTRA Edge Dashboard: Project details load faster; activity logs show more useful detail.
• SPEKTRA Edge Dashboard: More controls have tooltips; product naming in the UI uses SPEKTRA Edge.

2025-09-17

• AI: AI is available as a core platform service, including conversation logging with Azure OpenAI.
• Alerting: Alert evaluation supports function calling; timeseries conditions validate more accurately.
• Monitoring: Metric watches wait until data is ready, avoiding false gaps.
• Devices: Fewer false connectivity errors from proxies; device creation completes even when the record is created on first update.
• Logging: Log buckets remain usable when a descriptor reference is temporarily missing.

2025-08-27

• SPEKTRA Edge Dashboard: Notification channels let you choose which alert kinds are enabled.
• SPEKTRA Edge Dashboard: Alert edits save safely under concurrent changes; time ranges and pagination behave correctly.
• SPEKTRA Edge Dashboard: Pod AI alerting pages show breadcrumbs; service-account private keys match the expected format.

2025-07-23

• SPEKTRA Edge Dashboard: AI alerting filters and visuals are improved; alerting policies and conditions support document templates.
• SPEKTRA Edge Dashboard: Alerting policies can toggle notifications; names enforce length limits.
• SPEKTRA Edge Dashboard: Attestation domains and device updates are easier to manage; organization and project IP conditions are supported.
• SPEKTRA Edge Dashboard: Provisioning policy editing uses a tabbed layout; organization views show the correct project count.

2 - SPEKTRA Edge OS Release Notes

SPEKTRA Edge OS is released separately from the platform services, and each device runs the version you select for it. These notes describe what changed for device operators in each version, newest first. They cover externally visible behavior only; build tooling and test changes are left out.

Devices do not patch themselves. Each release is built against current upstream packages, including their security updates, so upgrading a device is how it receives those fixes. To move a device to a newer version, see OS upgrades.

2.3.4

Released 31 August 2026.

• Security: Disabling the SSH server now takes full effect. On a device where you had turned SSH access off, the device could still accept connections on port 22 and start the server again on the next connection, because the socket that listens on the server’s behalf was left running. Disabling now stops that listener as well as the server itself, and enabling restores it. Devices running an earlier 2.3.x version with SSH disabled are affected and remain reachable over SSH; upgrade them to 2.3.4 to apply the fix.

2.3.3

Released 11 June 2026.

• Security: Refreshed the upstream packages the image is built from, picking up the security updates published since 2.3.2.

2.3.2

Released 12 May 2026.

• Networking: Docker now follows the device’s proxy settings. The proxy is written into the Docker daemon configuration and reapplied whenever it changes, so image pulls go through it.
• Upgrades: Downloading an OS image honors the configured proxy. It previously ignored the proxy, so upgrades could not complete on devices that reach the internet only through one.
• Networking: Everything that makes outbound connections — OS image downloads, health checks, and external address detection — now uses the proxy the device actually connected through rather than a fixed setting. If the device falls back to a different proxy, or to a direct connection, the rest of the system follows.
• Security: Proxy credentials are masked in device logs.
• Security: Dependency updates addressing known CVEs in the Go standard library, golang.org/x/net, pion/dtls, and xmlquery.

2.3.1

Released 20 February 2026.

• Networking: A wrong proxy address no longer strands a device. Fallback now also tries the default proxy the image was built with, and it no longer deletes the active configuration — which previously left devices that need a proxy unable to reach the platform at all.
• Networking: Devices with more than one Ethernet port boot reliably. Boot now waits for any interface to come online rather than all of them.
• Networking: Repeated invalid network configurations no longer fill the configuration partition. Each failure overwrites a single debug file instead of leaving a new one behind.
• Workloads: Workloads start on boot before the device connects to the platform, so containers come up even while the device is offline.
• Logging: Container logs reach the Logging service again on devices set up to export specific systemd units. Container entries were being filtered out before export.
• Logging: Less noise in device logs from Ethernet and modem state reporting and from the status LED daemon.
• Security: USB control rules that leave fields unset now render correctly.
• Time: Time synchronization no longer drops its sources in response to network manager events, which had caused intermittent sync failures.
• Provisioning: A provisioning approval request carries the device’s status, so you can see what you are approving.
• Devices: Raspberry Pi devices report their manufacturer and product name on firmware that does not supply them through DMI.
• Devices: Boot no longer stalls waiting for input when a TPM or other HAT module is attached.
• Remote access: SSH, SCP, log streaming, and port forwarding moved to the current platform protocol, and an SSH session now reports its exit status exactly once.
• Platform support: Jetson Linux updated from 36.4.0 to 36.4.4.

2.3.0

Released 17 July 2025.

• Platform: The base system moved to a newer Linux release, bringing an updated kernel and refreshed system packages.
• Monitoring: Devices report network interface metrics.
• Monitoring: Container metrics are reported for containers whose names do not follow the default naming scheme.
• Workloads: Compose files support environment variable substitution.
• Workloads: Containers report a Restarting state.
• Workloads: Pod status is reported even when Docker is not running, rather than going unreported.
• Workloads: Unused Docker images are no longer pruned on every run. Pruning happens once a day, and only when disk usage is above 75%.
• Networking: Uplink fallback can be turned off for devices that should stay on their primary uplink.
• Time: Time synchronization switched from ntp to chrony.
• Reliability: The hardware watchdog is enabled where the hardware provides one, with a software watchdog as fallback, so a hung device restarts itself.
• Upgrades: OS image downloads report their size so progress is accurate, are cached and verified by checksum, and allow more time to finish on slow links.
• Provisioning: Fixed the manual approval flow for devices awaiting approval.
• Security: Devices no longer crash when the fail2ban configuration is missing.
• Security: Encrypted devices return to Started once decryption finishes, instead of staying in the decrypting state.
• Logging: Empty log batches are no longer sent to the Logging service.
• Devices: Kernel warnings no longer clutter the device console.
• Devices: Shutdown hooks run on Fitlet3 hardware, because shutdown now goes through systemd instead of bypassing it.
• ZTNA: WireGuard is installed on the device to support ZTNA connectivity.