> ## Documentation Index
> Fetch the complete documentation index at: https://docs.menaia.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Ai features and limits

> What Menaia's AI helpers do, the length and size limits on each one, and how every AI suggestion is reviewed before it's used.

This reference explains Menaia's AI-assisted features, the limits on what you can submit to each one, and the rules that keep AI output a draft you review rather than something applied automatically.

## What the AI features are

Menaia's AI helpers fall into three groups:

* **Meni, the AI assistant** — a chat panel that answers questions, looks up and charts your data, remembers context, prepares changes for your approval, and runs automations that draft work for you.
* **Drafting and writing helpers** that propose text you then review and edit — crew notes, the proposal email, text replies in the inbox, and win celebration posts.
* **Analysis helpers** that summarize a recorded sales conversation — ride-along insights.

### Nothing the AI produces is applied automatically

Every AI feature returns a **suggestion**. The result is never sent, saved, or applied on its own. The screen always drops the output into an editable field, an undo-able replacement, or a review step first. See [Always review AI output](#always-review-ai-output) below.

## What each feature does

### Meni chat

The Meni panel opens from the top bar on any page. With the full assistant, Meni answers how-to questions from these docs, looks up live records your role can see, and shows answers as charts, tables, KPI tiles, and clickable record lists. Without it, the panel runs in **Ask the docs** mode and answers only how-to questions and problem reports.

* Meni only uses data returned for your permissions, says where an answer came from, and doesn't invent names, numbers, or dates.
* Conversations with the full assistant are saved in **Conversation history**.

See [Ask Meni](/guides/meni/ask-meni).

### Meni memory

Meni can remember facts about leads and clients, and your own preferences, between conversations. Memories come from chats, follow-up emails sent to leads, and ride-alongs, and each one shows its source. You can forget any memory from the **Meni memory** panel. Meni treats a memory as recall and checks live data before acting on it.

See [Meni memory](/guides/meni/ask-meni#meni-memory).

### Meni actions

Meni can prepare changes — scheduling shifts, tasks, and time off; updating leads, jobs, clients, contacts, projects, job plans, and project comments; drafting follow-up emails and texts. Each one is an approval card, and nothing changes until you approve it. Meni can't take payroll, payment, invoice, user, or subcontractor actions.

See [Ask Meni to make a change](/guides/meni/ask-meni#ask-meni-to-make-a-change).

### Meni Dashboard

The **Dashboard** shows signals Meni spotted, your morning briefing, drafts waiting for approval, and live views you pinned from the chat. Pinned views refresh with the viewer's current permissions.

See [Meni Dashboard](/guides/meni/dashboard).

### Meni automations and daily insight

Playbooks react to business events and watch rules check your data on a schedule. Both draft into an approval queue — Meni drafts, never sends. Admins also receive a **Meni daily insight** email, which they can steer with **Context for Meni** under **Settings → Meni**.

See [Automations](/guides/meni/automations).

### Crew notes (Enhance / Suggest)

On the inspection-report step, the crew-notes field has an in-field AI button that rewrites your notes into a concise sales-to-crew heads-up.

* There are two modes: **Enhance with AI** when you've already typed notes, and **Suggest with AI** when the field is empty (it then drafts from the job context — customer demeanor, concerns, and special instructions).
* The result replaces the field but is **fully undo-able**. Your prior text is kept, and an Undo / Revert control restores it.

### Proposal email (Draft / Enhance)

On a proposal, the email body field has an AI button that rewrites the body into a warm, professional client-facing email (or writes a default body if the field is empty).

* The AI is told **not to invent pricing, dates, attachments, guarantees, or legal terms**, and to keep your factual claims intact.
* The result is undo-able — your prior body is kept before the new text is applied.

### Text reply

In the **Communications** inbox, when you're replying to a text conversation, the message composer has an AI button (the sparkles icon beside the message box), or you can type **/ai**. It's available once the conversation has started. The AI reads the recent back-and-forth of that conversation and drafts a short reply for you to send.

* While it works the button shows a spinner; the generated reply is then dropped into the composer where you can edit it before sending.
* The draft is placed in the composer as editable text — it is **never sent automatically**. Nothing goes out until you press send yourself.
* The AI is told to use **only the names, services, addresses, dates, and times that actually appear in the conversation** — it won't invent details or leave placeholders. If the customer asked something that can't be answered from the conversation, the draft acknowledges it and says you'll confirm.
* If any typed text is already in the composer, the draft is added at your cursor rather than replacing what you wrote.

### Win celebration post

In a Win group's feed composer, **Use AI** opens a three-step flow that drafts a celebration post from an existing win (a signed proposal, hours saved, or a badge unlocked) and tags the people involved.

* The steps are: 1) pick a recent win, 2) add an optional note or GIF, 3) **Review & Post** — the generated text lands in an editor so you can edit it before sharing.
* You can **Regenerate** on the review step if you don't like the draft.
* The AI is told **not to invent details that weren't provided**.

### Ride-along insights

The **AI Ride Along** recorder captures an in-person sales conversation. After you stop, the audio is transcribed and analyzed, and results appear as tabbed insights.

* The recorder explains: "Record your in-person sales conversation. When you're done, AI will analyze the conversation and provide coaching insights."
* The analysis is organized into sections: **Summary**, **Strengths**, **Next Steps** (or **Follow-up** if the deal closed on the spot), **Coaching** insights, **Tone** and communication analysis (talk-time percentage, confidence, engagement, hesitations), **Patterns**, and ready-to-send **Drafts** for a follow-up email or text.
* Per-speaker tone, confidence, and engagement come from voice analysis. When that voice data isn't available, those fields are shown as empty rather than guessed.

## Usage limits

### Length and size limits

| Feature | Limit |
| - | - |
| Meni pinned views | Up to 30 per person |
| Meni watch rules | Up to 20 per person; each checks between every 15 minutes and once a week |
| Meni playbook instructions | Up to 4,000 characters |
| Meni playbook daily draft cap | 1 to 100 drafts a day (20 by default) |
| Context for Meni (daily insight) | Up to 4,000 characters |
| Crew notes | Up to 5,000 characters (input is capped, and AI output is held to the same limit) |
| Proposal email body | Up to 6,000 characters (input capped, output held to the same limit) |
| Proposal email subject context | Up to 300 characters |
| Text reply draft | Up to 600 characters (a long draft is trimmed to fit a text) |
| Win-post note | Up to 500 characters |
| Ride-along recording | Auto-stops at 2 hours |

### Processing time for ride-alongs

* Ride-along analysis "usually takes 30 to 90 seconds. You can navigate away — the analysis will continue in the background."
* The results screen checks for completion automatically and stops checking after roughly 15 minutes, then shows "Still processing — check back soon … Refresh the page in a few minutes to check on the analysis."
* If analysis can't complete you'll see **Analysis Failed** with the reason, or "An unexpected error occurred during analysis."

## Meni tokens power AI features

On workspaces that use token billing, Meni and every AI feature draw from a shared workspace **token balance**. Each AI action spends tokens from that balance — as do outbound calls and texts — and the activity shows up in your **Usage history** under **Settings → Billing**.

* Your plan includes a monthly token allowance, and a **Trial** workspace starts with a pool of trial tokens.
* When the balance runs low you'll see a warning; when it reaches **zero**, Meni, the AI drafting and analysis helpers, and outbound calling and texting **pause** until you top up. Your core business tools — scheduling, clients, proposals, invoices, and everything else — keep working and never pause.
* If an AI action is refused because the balance is empty, you'll see a message that your workspace is out of Meni tokens and needs more to keep using AI features and outbound calling and texting.
* Admins top up under **Settings → Billing** — either **Buy tokens** or by turning on automatic token reloads (auto-reload). Not every workspace uses token billing; where it isn't in use, AI features aren't limited by a token balance.

## Who can use each AI feature

AI features are turned on per workspace and, in some cases, depend on your role, so not every workspace or user sees them.

* **Meni** — everyone can use **Ask the docs**. The full assistant, Meni memory, automations, and workspace data views are each turned on separately. The **Dashboard** is for Admins, Ops Managers, Crew Leaders, and Crew Members (and Sales Admins and Sales Members when the sales overview is on); **Automations** is for Admins, Ops Managers, Crew Leaders, and Crew Members; the daily insight email goes to Admins. See [Feature availability](/reference/platform/feature-availability#meni-assistant).
* **Crew notes and the proposal email helper** appear where you can edit the underlying record — the inspection-report step and the proposal email, respectively.
* **The text reply helper** appears in the inbox composer for people who can send texts, and only once texting has been turned on for your workspace.
* **Ride-Alongs** must be turned on for your workspace. The voice tone analysis depends on an additional voice-analysis setting being turned on for your workspace.
* **AI win posts** require win channels to be turned on. With them off, you'll see "This feature is not available for your organization."

If an AI feature you expect is missing, it's almost always because the feature isn't turned on for your workspace, or your role can't edit the underlying record. If an AI feature is available but suddenly refuses to run, check your **token balance** — AI features pause when it reaches zero. See [Permissions and roles](/reference/platform/permissions-roles-matrix).

## Always review AI output

AI output is a draft to check, not a finished artifact — and this is enforced by the interface, not just advised.

### Meni changes wait for your approval

Every change Meni prepares — in the chat or from a playbook — is an approval card or a queued draft. Meni re-checks your permissions and the current record when you approve, and refuses if the record changed in the meantime. Approving a follow-up email from the automation queue asks for a second confirmation because it reaches the customer right away.

### Text replacements are undo-able

Crew-notes and proposal-email replacements can be reverted (Keep / Revert), so you can compare against your original and roll back.

### Win posts open in a review step

A win post opens in an editor on a dedicated **Review & Post** step (with **Regenerate**) before anything is shared to the feed.

### The AI is constrained against making things up

* Proposal emails must **not invent pricing, dates, guarantees, or legal terms**.
* Win posts must **not invent details** that weren't provided.
* Text replies must use **only the names, services, addresses, dates, and times that appear in the conversation**.
* Meni must answer only from data it looked up for you, and must **never claim a change happened** unless it was approved and completed.

### Customer-typed text is treated as untrusted

All AI features treat user-typed text as content to summarize, not as instructions to follow — so odd phrasing in a customer's notes can't redirect the AI. This is worth knowing if a customer's notes contain unusual wording.

## Related references

* [Permissions and roles](/reference/platform/permissions-roles-matrix)
* [Feature availability](/reference/platform/feature-availability)