Sprout
The assistant in your Graine dashboard. Ask in your own words; Sprout reads what it needs and suggests each change as a card, before → after. Nothing changes until you approve, and it works with your own permissions and your workspace's approval rules.
Sprout is the assistant built into the Graine dashboard. Tell it what you want in your own words — "make this agent's greeting friendlier", "build a workflow that calls new leads and tries twice more if nobody answers", "pause the Diwali campaign" — and it reads what it needs, answers, and suggests the changes as cards. Nothing changes until you approve.
Sprout acts as you: it can see and do only what your role allows, and your workspace's approval rules apply to it exactly as they apply to you.
Opening Sprout
- Ask Sprout at the top of the sidebar, on every page of the dashboard. When the sidebar is closed, the Sprout button is at the top of the page.
- ⌘J on a Mac, Ctrl+J on Windows and Linux. Press it again inside Sprout to close it, or press Escape.
The panel opens on the right — beside the page on a wide screen, over it on a narrower one, and full screen on a phone. Sprout knows the page you are on: on an agent's page, "this agent" means that agent; on a workflow, a campaign or a call, the same.
What Sprout can do
| Area | Sprout can |
|---|---|
| Agents | List and read your agents; create one as a draft, from a few words or from a brief or SOP; change its name, description, welcome message and prompt, its AI model and how long its replies are, its voice's speed, or give it another agent's voice; switch it live or off; make a copy; place a test call to a number you give |
| Workflows | List and read your workflows and the samples; check a workflow for problems; create one; change its draft; publish it; test-run it; follow a run, and cancel one |
| Campaigns | List your campaigns and see how one is doing; start, pause and resume a campaign, or a workflow's list |
| Calls | Find calls and read one, with its transcript |
| Phone numbers | List your numbers, and which agent answers each; assign a number to an agent |
| Your workspace | Read your balance; list your knowledge bases; answer questions about Graine from these docs |
| Pages | Open the page you need |
Sprout offers only what your role allows. On a View only seat it reads and answers, and suggests nothing that changes anything (Team roles).
How approval works
Every change Sprout suggests arrives as a card:
- What it changes, and on which agent, workflow, campaign or number.
- Every field before → after. Long text folds; open it to read all of it.
- A warning on anything that reaches the outside world. Places a real call marks a test call, a test run, and starting or resuming a campaign or a list (Places real calls). Changes something live marks the rest that act on what is running now — publishing a workflow, pausing a campaign.
Approve carries it out; Discard changes nothing. A card with several changes has a checkbox on each, so you can approve only some of them. You have 15 minutes to decide — the card counts down — after which the suggestion expires and nothing is changed; ask Sprout again and it suggests it afresh. Only you can approve what Sprout suggested to you.
Once you approve, each change shows how it went:
| Result | What happened |
|---|---|
| Done | The change was made. Open takes you to it. |
| Sent for approval | Your workspace sends this kind of change to an admin first; it waits in Approvals, as it would if you had made it yourself. |
| You don't have permission | Your role can't make this change. The card says why. |
| Failed | It couldn't be made. The card says why. |
| Changed since this was suggested | Someone changed it after Sprout read it. Ask again and Sprout starts from what is there now. |
Sprout then carries on: it tells you what it did, and may suggest what comes next. Pages showing what changed — a workflow, the lists of agents, campaigns and numbers — update by themselves.
Reload the agent editor after a change
When Sprout changes the agent whose page is open, Sprout asks you to reload. The editor still shows the settings from before, and saving it without reloading would put them back.
Undo a change
When a change can be put back, its result shows Revert. Sprout then suggests the change that undoes it — before → after, on a new card — and nothing is put back until you approve that card too. A change that cannot simply be undone, such as a call that was placed, doesn't offer Revert.
Create an agent from a brief or an SOP
Choose Create an agent with Sprout on the Agents page, or in an empty Sprout panel. Paste a brief — who the agent calls, why, what it should say and find out — or attach the document itself, and choose Draft the agent.
| Files Sprout reads | |
|---|---|
| Kinds | PDF, Word (.docx) and plain text (.txt) |
| Size | Up to 10 MB, one file at a time |
| How | Drop it on the panel, or choose it — in the brief box or next to the message box |
| Long documents | Sprout reads the first part (about 40,000 characters) and says Only the first part was read |
A file Sprout can't read is refused in plain words: one over 10 MB, one of another kind, one protected with a password, and one with no text in it — a scan saved as an image, say. Save a copy that has text, without a password, and attach that.
What comes back is a card for a new draft agent: its name, its welcome message, the voice and speech recognition it will use, and its prompt in three blocks, each a section you can open and read in full:
| Block | What it says |
|---|---|
| Who the agent is | Its name, your company, its role and tone, and who it talks to |
| The conversation | The goal, the steps, the questions to ask, what never to do, and how to answer objections |
| How the call ends | The closing line, when to hang up — interested, busy, not interested, a wrong number — and what to collect before it does |
Before it is suggested, the prompt goes through a 10-point check: identity, goal, tone, greeting, conversation steps, guardrails, objection handling, closing line, when to hang up, and whether the language and the voice fit. The card lists what needs attention first, then what passed. Ask Sprout to fix what it flags before you approve — "add what to say when they say it's too expensive" — and it suggests the changed block, before → after.
The agent is created as a draft. Test it and put it live as you would any other agent.
Your conversations
- Recent conversations (the clock at the top of the panel) lists your 30 most recent. Open one to carry on where you left off.
- On an agent's page, Sprout carries on with your latest conversation about that agent, so what you asked about it last time is still there.
- New chat starts afresh. Deleting a conversation removes it; changes you approved in it stay as they are.
- If your connection drops while Sprout is answering, it carries on. Press Retry (or open the conversation again) to see its answer — a question it already has is never sent twice.
What Sprout can't do yet
- Delete or archive anything: agents, workflows, campaigns, lists or numbers.
- Change an agent's languages, its voice provider, speech recognition, the knowledge and tools it uses, or its calling hours — Sprout offers to open the agent editor for those.
- Buy, release or port numbers, or change SIP trunks and carrier settings.
- Add credits, or anything about invoices and plans — it can read your balance.
- Change your team, roles, API keys, integrations, webhooks, compliance settings or sub-accounts.
- Add or remove knowledge base documents, voices in your voice library, web components or the website embed.
- Approve a request on someone else's behalf, or carry out a change you haven't approved.
Privacy, in plain terms
- Sprout sees what you can see. It reads the dashboard through your own access — the agents and data your role reaches, and nothing more.
- It is told the page you are on: the page's address and the ids in it. Not what is on the page, and not what you typed into a search box.
- An attached file isn't kept. Only its text is, for that conversation.
- Your conversations are yours. No one else in your workspace can see them. They are kept for 30 days after you last use one, and you can delete one at any time.
- Sprout's answers are written by an AI model. Read each change on its card before you approve it — that is what the card is for.
Limits
| A message | Up to 4,000 characters; a pasted brief, up to about 3,800 |
| Messages | 30 per person every 10 minutes, and 300 per workspace an hour |
| Attached files | 10 per person every 10 minutes |
| Answers | One at a time in each conversation |
At a limit, Sprout says how long until you can send again.
The same tools in your AI assistant
Sprout's tools are also an MCP server, for Claude, Cursor and other AI tools you use: see MCP server.

