Release History

0.5.0 - 2026-10-14

New features

Device integration and modeling

• IoT platform integration: Adds EMQX Fleets and ThingsNeural integrations using each platform's native device models, identities, and operations. See IoT Platform Integration.
• Modeling and simulation: Adds energy-storage and PV-inverter modeling references and previews of how commands affect simulated state. SDKs and simulators follow the connected platform's native model and protocol.
• Device history queries: ThingsNeural supports time-bounded property and event history queries, with statistics grouped into time intervals.

Conversations and voice

• Agent configuration: Adds Configuration Overview for conversation rules, recognition hotwords, voice tool scope, responses, and feedback.
• Model reasoning: Supports enabling reasoning, selecting model-supported effort levels, and viewing thinking content returned by the model in chat.
• Long-term memory and context: Supports saving, inspecting, and deleting long-term preferences. Chat displays context usage and reports automatic context compaction. See Conversations and Memory.
• Voice feedback: Supports categorized waiting messages, question matching, query preloading, audio reuse, and response timeout settings.

Workflows and multi-agent collaboration

• Workflow editor: Adds a visual editor with Agent tool and skill selection, IM notifications, device state updates, and device job actions.
• Jev decision model: Adds workflow decision nodes that select a branch from an event and predefined options, including an insufficient-evidence branch. Standalone deployments also support device quick commands.
• A2A collaboration in chat: Supports delegating tasks from text or voice conversations to agents for bound devices and returning results in the original conversation. Controller-device bindings define the allowed targets. See A2A Multi-Agent Collaboration.
• Scheduled device reminders: Scheduled tasks can send due reminders to MQTT devices within their scope that support receiving reminders.

Enterprise Platform

• Production deployment: Provides a commercial Platform for deploying, scaling, and centrally managing Device Agents on enterprise-owned infrastructure.
• Runtime monitoring: Adds monitoring for platform operation, connections and sessions, task outcomes, and response latency, with per-device AI usage, estimated costs, cumulative usage, and voice interaction metrics.
• Device AI credits: Supports CNY or USD pricing, independent device budgets, charging rules, and device balance inspection and adjustment. See AI Credits.
• License authorization: Supports signed license files that define Agent requests per minute (RPM) and validity. See License Management.

Improvements and fixes

• Shows desired state, state differences, enum descriptions, and command history and device job records provided by the connected platform in the device state panel.
• Fixes lost local configuration during ThingsNeural synchronization and shows synchronization failure reasons.
• ThingsNeural fills in connection settings after login authorization and guides rebinding when credentials expire.
• Improves browser simulator connection persistence, reconnection, and parameter-mapping compatibility.
• Refreshes published A2A capability cards after model saves or platform synchronization.
• Adds the Doubao Ark model provider, improves Cloudflare configuration, and updates DeepSeek Flash and Kimi model selection and reasoning options.
• Extends Aliyun speech-recognition and speech-synthesis models and their voices.
• Improves recovery from voice tool errors, answer context, and interruption handling to reduce interference from previous voice turns.
• Improves mobile workspaces and chat layout, history pagination, and response-language matching. Clearing a conversation archives its history and retains long-term memory.
• Fixes session history loss after starting the service from a different directory.
• Fixes time handling in device state and event queries: times are parsed in the server time zone, returned as full UTC datetimes, and latest-event queries no longer use an empty time range.
• Updating device state and calling device jobs no longer require approval on the Platform; duplicate execution is still prevented.
• Shows validation errors when importing a skill archive.
• Provides Docker image installation.

0.4.0 - 2026-07-25

Before upgrading

• When multiple Gateways share one database, stop every 0.3.x Gateway before starting 0.4.0. Do not mix both versions during a rolling update, or a scheduled task may run more than once.
• Update bookmarks and integrations from /workspace/marketplace to /workspace/a2a; the old route is not redirected. Remove the obsolete VITE_FF_A2A_MARKETPLACE_ENABLE setting.

New features

• Adds built-in multi-agent orchestration for collaborative Tasks and Scenes, including unified conversations, final answers, device binding, Scene management APIs, and SSE execution APIs.
• Adds reliable scheduled task execution with a dedicated Schedules page, persisted execution context, run history, and detailed run results.
• Adds reusable outbound Webhooks with connection management, delivery testing, and workflow integration.
• Adds direct HTTPS for the Device Agent console and HTTP API.

Improvements and fixes

• Replaces the former Marketplace workspace with A2A and improves task and scene interactions.
• Improves chat composer behavior for multi-line input.
• Cleans up generated device SDK toolkit bundles.

0.3.0 - 2026-07-01

New features

• Adds workflows for device reports, commands, MQTT messages, and Agent handoff.
• Adds a Workflows page with graph view, controls, fullscreen view, and run history.
• Adds branching workflow nodes for multi-path automation.
• Adds simulator event reports with editable payloads.
• Adds Device Agent persona settings for reply style.
• Adds Vehicle examples based on VSS.

Improvements and fixes

• Improves Logs page layout and split-pane persistence.
• Updates the model catalog in workspace settings.
• Improves voice and MQTT connection details for remote demos.
• Fixes browser simulator registration timing.
• Improves command targeting when no product context is selected.

0.2.0 - 2026-06-09

New features

• Adds enum support in DeviceSpec for fixed-value device modes and status fields.

Improvements and fixes

• Enforces product-scoped tool and device access so a Device Agent only operates within its own device model.
• Renames telemetry to properties across the device model, UI, and API documentation.
• Removes legacy non-console routes and streamlines console navigation.
• Improves DeviceSpec preview, sidebar name truncation, and embedded device spec extraction for long fields and long agent names.
• Fixes LAN access for the development server, IM approval handling, and vertical wrapping for property names.

0.1.0 - 2026-05-20

• Adds the About & Contact page and simplifies workspace navigation with a shared sidebar pattern.
• Adds the EMQX Cloud entry point in workspace headers and keeps language labels stable as 中文 / English.
• Improves MQTT runtime configuration, including Zero EMQX provisioning, MQTT broker settings, model settings, and voice settings documentation.

0.0.8 - 2026-05-19

• Shows Zero EMQX setup errors directly in the MQTT settings dialog so you can see why setup failed and retry.
• Flattens the installation documentation route from /installation/introduction to /installation, with navigation links updated.