Assistant
AssistantAmbiguous Assistant. An AI teammate that plans, executes, and self-corrects across your workspace.
Ask Ambi to do something and it actually does it. The built-in Assistant plans what needs to happen, works across Docs, Mail, CRM, Calendar, and every other app in your workspace, then checks its own work before handing it back. It thinks like a teammate, not a chatbot.
How humans use the Assistant
Open Assistant from any page in your workspace. Type a prompt or pick a suggested action. Watch the SPEAR phases run: understanding, planning, executing across modules, assessing quality, resolving with a deliverable. The output lands as a real artifact (doc, deck, email draft, task list) you can edit, share, or send.
Transparent phases
Watch each SPEAR phase in real time. See what the Assistant is planning, which tools it invokes, and what it self-corrects before delivery.
Included from day one
The rest of the Assistant feature set. Every item ships in every workspace, for your team and your AI coworkers alike.
Natural-language prompts
Type what you need done in plain language. The Assistant scopes the work, identifies which modules to use, and builds a plan before executing.
Cross-module execution
One prompt can create a doc, send an email, update the CRM, and schedule a meeting. The Assistant uses every module as a tool, not just one.
Self-correction (Assess phase)
After executing, the Assistant checks its output against the original prompt. If quality falls short, it re-runs the relevant steps before delivering.
Artifact output
Results land as real workspace objects: a Slides deck, a Docs draft, a Tasks list. Not just chat text. Edit them like any other file.
Prompt suggestions
Context-aware chips suggest prompts based on your current page and recent activity. One click starts a multi-step workflow.
One product, two ways in
External coworkers call the same Assistant surface via MCP or REST. The API accepts a prompt and streams SPEAR phases back as SSE events. Your agent can run the Assistant as a sub-tool, delegation within delegation.
// Run the Assistant via MCP
const result = await mcp.callTool("assistant.chat", {
prompt: "Summarize the last 5 emails and create a doc with action items.",
max_rounds: 5
});
// Returns: { phases: [...], output: { type: "doc", id, url } }What your agent can do in Assistant
Every action your team takes in Assistant, your agent can do the same way.
| Method | Description |
|---|---|
| assistant.chat | Send a prompt and get a planned, executed response |
| assistant.chatStream | Stream a response with phase-by-phase progress |
| assistant.capabilities | List all tools and modules the assistant can use |
| assistant.recentActivity | List recent actions and tool calls |
| assistant.replay | Replay a previous assistant conversation turn |
Not a chatbot sidebar
The Assistant is a full AI coworker running on the SPEAR loop: Scope, Plan, Execute, Assess, Resolve. It reads your prompt, breaks it into steps, uses every module in your workspace (Docs, Mail, CRM, Calendar, Slides), checks its own work, and delivers a finished artifact. Same surface for humans typing prompts and for external agents calling the API.
Get started
Frequently asked questions
The Assistant has tools across all 17 modules: create docs, send emails, update CRM deals, book calendar events, manage tasks, and more. It plans, executes, and self-corrects using the SPEAR loop.
Each Assistant tool call counts as one AI action against your monthly budget (1,000 free actions on the free plan). Premium operations like image generation cost more per call and show a badge before you confirm.
