V5.7.0

Released November 2, 2026

Native Connectors, Neural Memory & Staged Startup

Symbiote 5.7.0 adds a first-class connector platform, agent-enabled Markets and intelligence workspaces, optional local neural memory, and a more responsive startup and recovery experience. It also brings major improvements to hands-free voice, the agent's state animations, and the Memory graph.

Major upgrades

• Faster, staged startup: The Gateway now reports core readiness before launching integrations, channels, memory, market-data, and self-improvement initialization as concurrent Gamma background tasks. Optional startup work no longer holds the core ready signal hostage. Electron displays a dedicated animated thinking-orb startup screen, with a clear status if the Gateway takes longer to connect. Heavy webapp views are loaded on demand.
• A coherent four-state agent orb: Replaced the scattered particle swarm with a single, deterministic hero animation that transitions between idle, thinking, executing, and responding states. The animation uses 1,000 particles and reflects the live agent state.
• Connector platform: Added manifest-based REST and MCP adapters, encrypted credential storage, explicit per-agent capability grants, connection health checks, and bounded execution history. The initial REST catalog includes GitHub, Notion, Slack, Figma, and Canva.
• Markets and public intelligence: Added an agent-enabled Markets workspace with public price and history feeds, persistent chart drawings and indicators, a chart tool, and dedicated chat and voice input. An optional MetaTrader 5 bridge supports market data and trade-journal synchronization. The resource aggregator adds OpenSky aircraft positions and further public market and intelligence sources.
• A richer Memory experience: Added optional local neural retrieval using Ollama embeddings and a persistent HNSW index alongside lexical VDB search. Restored the Memory map as a responsive 3D force graph, with clustered nodes, filtering, node details, and image export.
• Hands-free voice: Added profile-driven wake-word activation so the phrase follows the active agent's name rather than a hardcoded alias. Voice activation can open a fresh chat with listening enabled; dictation supports a trailing “send” command. Added optional offline Vosk speech recognition with a persistent worker.
• Provider and workflow options: Registered FreeAI in the CLI and Gateway and added provider/model discovery for interactive switching. Added admin-only n8n workflow operations and safer local-first integration controls.

Fixes and reliability

• Improved speech recognition for short recordings by finalizing audio before recognition and adding trailing silence. Wake-word matching now normalizes names and tolerates common speech-recognition variation. Offline Vosk falls back to Windows speech recognition when its package or model is unavailable.
• Improved Gateway startup and recovery reporting so background initialization failures are visible and do not prevent the core-ready state. Electron supervises Gateway startup and provides a useful recovery status rather than leaving an uninformative blank window.
• Improved chart-state persistence and public data-source status handling.
• Corrected the hero orb's display size to prevent stretching and removed its brightness control without affecting global appearance settings.
• Improved Windows project-terminal lifecycle handling and session continuity.

Additional improvements

• Refined the API registry, integrations, browser, IDE, maps, health, and Markets interfaces for more consistent layout and theme behavior.
• Added CI, CodeQL, release publishing, dependency-update, and repository-label automation.
• Added desktop and mobile screenshots for the webapp, native recovery UI, and CLI.
• Updated release packaging to publish version-matched assets and SPDX software bills of materials.

Upgrade notes

• No existing configuration migration is required.
• Connectors are optional. Provide credentials for each service you choose to connect and grant agents only the capabilities they need.
• Neural memory is optional and requires Ollama plus a locally installed embedding model. The example configuration uses bge-m3; qwen3-embedding:8b is another option. Existing lexical memories remain available without neural memory.
• Offline Vosk recognition requires the vosk Python package and the vosk-model-small-en-us-0.15 model under .symbiote/models. Set SYMBIOTE_VOICE_MODEL to use a model elsewhere. Windows speech recognition remains the fallback if Vosk is unavailable.
• The MetaTrader 5 bridge is optional and requires a configured MetaTrader 5 terminal and its Python integration.
• To build from a source checkout, run npm ci, npm --prefix webapp ci, then npm run build.