AI Assistant: workspace chat, token budget and local speech recognition
The AI Assistant is a chat with access to your workspace (tasks, meetings, projects), including conversation history and a token-budget display. Separately, "Voice" is a fully local dictation tool with no server connection at all.
- You're signed in and your organization has the "AI Assistant" module activated.
- Chat requests consume a pool of AI tokens whose limit depends on the subscription plan.
- 1
Manage conversations in the sidebar
The sidebar on the left lists every past conversation; "New" starts a fresh one, clicking an existing one loads its history. The trash icon (on hover over an entry) deletes a conversation permanently, with no confirmation. The sidebar toggle at the top can hide the list entirely.
- 2
Use suggestions and workspace context
A new conversation shows three suggestion chips ("What are our next steps?", "Create task: …", "What are my open tasks?"). The assistant knows your workspace — it can create tasks directly or look up open items across projects and meetings without you pasting in context manually.
- 3
Keep an eye on the token budget
A tile at the bottom of the sidebar shows AI token usage against your plan's limit, with a "More tokens" link to billing. Going over the limit doesn't cut you off immediately — a grace buffer ("overage") applies and gets billed on the next invoice instead.
- 4
Dictate via microphone instead of typing
The microphone icon in the input field uses the browser's own speech recognition (Web Speech API) for live dictation straight into the chat text field — it only works in browsers that support it (mainly Chrome-based) and sends audio to the browser's own recognition, not to a dedicated transcription service.
- 5
"Voice": fully local transcription, no server involved
Under AI Assistant → "Voice" (a separate page) sits a completely different tool: recording and transcription run entirely in the browser via Whisper (Transformers.js) — no audio is ever sent to a server. It's suited for longer dictation with zero token usage, but it isn't a direct chat with the assistant.
Deleting a conversation doesn't ask for confirmation
The trash icon in the sidebar removes a conversation immediately and permanently — no confirmation dialog and no undo.
Two separate speech paths — don't mix them up
The microphone in the chat input transcribes live into the message via the browser's built-in speech recognition; the separate "Voice" page is a standalone, fully local Whisper tool with no connection to chat history. Neither shares recordings or results with the other.
Missing provider approval shows a generic message
If AI model access with the provider (Anthropic/Bedrock) hasn't been approved yet for this deployment, you get a notice that access is "not yet enabled" — resending won't fix it; it requires approval from an admin/operator.
Overage gets billed, not blocked instantly
Going over the token limit keeps chat working via a grace buffer at first — the extra usage shows up on the next invoice instead, which isn't obvious at a glance.
More articles are coming soon.