Wave Assistant
From Wave Wiki — the open reference for the Wave OS ecosystem.
1. About
Wave Assistant is the built-in AI assistant for Wave OS. It lives across the entire operating system, with context awareness of your files, wallet, compute sessions, and connected apps. Unlike a standalone chatbot, Wave Assistant can read from and act on your Wave OS data — create calendar events, search notes, manage files, answer questions about your projects, and route requests to the appropriate Wave OS app. It is powered by Theta EdgeCloud's GLM-5.2 model, with fallback to other models via BYOK (Bring Your Own Key).
2. Tech Stack
AI Model: GLM-5.2 via Theta EdgeCloud (primary), with BYOK OpenAI/Anthropic as fallback. Backend: waveChat.ts handles chat, 5 memory actions, an intent parser, and AI routing.
Memory: the AssistantMemory entity stores saved memories with auto-categorization. Context: a two-layer system prompt combines a shared base prompt with a team-specific overlay.
Sub-agents: a maximum of 5 run in parallel, with a cap of 20 per session. Knowledge: the KnowledgeBase entity organizes reference material into categories — dApp, Tutorial, and Reference.
3. Install
Wave Assistant is a built-in Wave OS application. No installation required — accessible from the desktop taskbar. Requires a Wave OS account. BYOK keys can be configured in Settings → Connectors.
4. Pricing & Credits
Free tier uses Wave OS credits — chat costs 1 credit per message. Memory operations (saveMemory, recallMemory, and other memory actions) cost 0 credits. BYOK lets users bring their own API key to bypass Wave OS credit usage. Theta EdgeCloud calls use platform Theta API tokens, with no additional cost beyond credits.
5. Known Limitations
No streaming responses — Wave Assistant uses a request-response model via Theta EdgeCloud. Sub-agent usage is capped at 5 parallel, 20 per session. Memory context window is limited to AssistantMemory records. There is no model selection field in the Theta API — the model is determined by chatbot configuration. Voice activation requires microphone permission.
6. Privacy & Data
Chat messages are processed through Theta EdgeCloud and are not permanently stored unless explicitly saved to memory. AssistantMemory records are stored in the Wave OS database with row-level security. KnowledgeBase entries are workspace-scoped — visible to all workspace members. BYOK API keys are encrypted with AES-256-GCM and never exposed to the client. Memory auto-categorization runs locally.
7. Published
Published: Yes. Part of the Wave OS main release, by xBuildy.