Assistant
The in-dashboard AI assistant — chat about your account, build and edit flows, and get help with any part of the platform.
The Talkif Assistant is an AI chat built into the dashboard. It answers questions about your account, proposes edits to your flows, and can build a complete flow from a plain-language description. It is available on every dashboard page and adapts to the page you're on.
What it can do
| Capability | How it works |
|---|---|
| Answer questions about your account | The assistant can look up your account data — flows, calls, campaigns, contacts, and more — with read-only access scoped to your account |
| Edit flows | On flow pages, it proposes complete flow changes as artifacts with a diff against your current definition. You apply or reject — the assistant never changes anything on its own |
| Build a complete flow | Describe what the flow should do and the assistant designs, writes, and creates a ready-to-edit draft flow — from any page, no empty canvas required |
| Propose actions | The assistant can suggest platform actions (as confirm buttons in the chat). Nothing executes until you confirm — every action runs under your own session |
| Escalate to support | The support side of the assistant can hand off to the Talkif team. On confirmation, your message is sent to support with your account and email attached, and the team follows up by email |
The assistant is routed to a specialist for the page you're on — flow design, call analysis, outreach, support, or general platform guidance. Routing is automatic; you just chat.
Flow builds
Ask for a new flow — for example, "Build a receptionist flow for my dental clinic" — and optionally give your business website URL. The build runs server-side through four steps:
| Step | What happens |
|---|---|
| Research | Investigates your business website (skipped when no URL is given) |
| Architect | Designs the node structure, transitions, and services |
| Write prompts | Writes each agent node's prompt, one node at a time, with a critic reviewing quality |
| Assemble | Validates the result and creates the draft flow |
Progress streams to the dashboard live over the flow-build-progress event, including per-node prompt-writing status. When the build completes, you get a normal draft flow — review and refine it in the Flow Builder before publishing.
One flow build runs per account at a time; starting a second while one is running returns a conflict.
Editing existing flows
On a flow page, the assistant reads your current definition and proposes changes as an artifact — a complete updated flow with validation results and a visual diff. Apply it to update the canvas, or keep chatting to iterate. Proposals are validated before they reach you, and publish-level issues surface as warnings.
Conversations
Conversations are saved server-side. You can list your recent conversations (up to 50) and reopen any of them with the full transcript. Conversations are private to you within the account — other members of the same account never see them.
If a conversation grows too long for a single turn, the assistant tells you to start a new chat.
Usage & billing
Assistant usage is metered by model token spend and drawn from three sources, in order:
| Tier | What it is |
|---|---|
| Monthly allowance | A free monthly allowance, reset at the start of each UTC calendar month with no rollover. Active once your account has a verified phone and a card on file |
| Bonus credit | 10% of every real balance top-up is granted as bonus assistant credit, valid for 90 days. Consumed after the monthly allowance, oldest expiry first |
| Account balance | Once the free tiers are exhausted, further usage is billed to your credit balance and appears as Assistant Usage in your transaction history |
The dashboard shows your current tiers, active bonus grants, and the next reset date.
Limits
| Limit | Value |
|---|---|
| Prompts per account | 10 per minute |
| Conversation history list | 50 most recent |
| Flow build request length | 10–4,000 characters |
| Concurrent flow builds | 1 per account |
When something goes wrong mid-chat, the error message states the specific cause (for example, a conversation that outgrew its context window) and includes a request reference you can quote to support.