AI Features

How Muin's AI works across documents, forms, and finance — what it automates, how accurate it is, and how your data stays private, never training public models.

17 questions

How does the AI keep my data private?

Muin takes your data privacy seriously. Here's how we protect your information: **Enterprise-Grade AI Infrastructure:** Muin's AI runs on AWS Bedrock — AWS's managed foundation-model service for enterprise AI workloads. Per AWS's published data privacy commitments: - Your data is **never used to train AI models** - Your data is **never retained** after processing - Complete isolation between customers **Encryption:** - All data encrypted in transit (TLS) - All data encrypted at rest (AES-256) - Field-level encryption for sensitive PII (AES-256-GCM) **Compliance:** - SOC 2 Type II: Planned (architecture designed for compliance) - HIPAA: Designed for compliance (BAAs not currently offered) - GDPR: Designed for compliance - Full audit logging **Your Control:** - You can delete your data at any time - Data is never sold or shared for advertising. Processing occurs through vetted sub-processors listed in our [DPA](/legal/dpa) - No training on your documents without explicit consent

Why did Muin categorize my document incorrectly?

AI categorization is based on analyzing document content, but it's not perfect. Here's what to do: **To correct the category:** 1. Open the document 2. Click **Edit Details** or the category badge 3. Select the correct document type 4. Save your changes **Why this happens:** - Unusual document layouts - Mixed content types (e.g., invoice with contract terms) - Low-quality scans - Documents in uncommon formats **Your corrections help!** When you correct a categorization, Muin learns from it. Over time, similar documents will be categorized more accurately. **Consistently problematic?** If certain document types are always miscategorized, contact support. We may be able to improve our models for your use case.

What can I ask the AI assistant?

Ask Muin can help you with a wide range of tasks: **Finding Information:** - "Show me invoices from last month" - "Find contracts expiring in 30 days" - "What policies need acknowledgment?" **Understanding Documents:** - "Summarize this contract" - "What are the key terms in this agreement?" - "Extract payment terms from this invoice" **Getting Help:** - "How do I upload a document?" - "What does this error mean?" - "Help me set up a new employee" **Taking Action:** - "Create a task to review this document" - "Send this invoice for approval" - "Remind me about contract renewal in 30 days" **Tips for better results:** - Be specific: "Invoices over $1000 from Acme Corp" vs just "invoices" - Muin knows your current page context - Ask follow-up questions naturally

What does the Usage Analytics dashboard show?

The Analytics dashboard (**Analytics** in the sidebar, under Intelligence) gives owners and admins a workspace-wide view of how Muin is being used: - **Agent performance** — runs, success rate, and average completion time for AI agents - **Workflow completion** — how many automations ran and finished successfully - **Token usage** — LLM tokens consumed across all agent runs - **Feature adoption** — which modules and features your team actually uses Use the period selector (7, 30, 90, or 365 days) to compare activity over time.

Why is my Analytics dashboard empty?

Analytics are computed from your workspace's actual activity, so a new workspace starts empty. Data appears as your team: 1. Runs AI agents or chats with the assistant 2. Executes workflows 3. Uses platform features day-to-day If you expect data but see none, check the period selector — a 7-day window on a quiet week can be empty while the 90-day view has plenty. Data refreshes automatically; there is nothing to configure.

Who can see the Usage Analytics dashboard?

The Analytics dashboard is limited to **owners and admins** — it shows workspace-wide operational metrics that aren't relevant (or appropriate) for every member role. If a team member needs analytics access, an owner can change their role in **Settings** → **Team**. For business reporting that anyone can run (financial statements, vendor spend, donor reports), use the **Reports** module instead.

Can I talk to Muin instead of typing?

Yes — in **Chat**, click the microphone button next to the message box, speak, and stop the recording. Your words are transcribed automatically and the assistant responds as if you'd typed them. Short, action-shaped phrases (like dictating an expense) are recognized as quick commands; everything else flows as normal conversation. Muin auto-detects which you meant.

Does voice input work in languages other than English?

Yes — transcription supports multiple languages, so speak in the language you work in. For best accuracy in any language: minimize background noise, speak naturally, and keep clips to a sentence or two. You can always edit the transcription before sending if a name or organization-specific term came through imperfectly.

What's the difference between voice input and the voice briefing?

They're opposite directions. **Voice input** is you talking to Muin — recording in chat and having it transcribed. The **voice briefing** is Muin talking to you — your daily briefing narrated as a podcast-style audio episode you can play in-app or subscribe to via a private RSS feed. Voice input lives in **Chat**; the voice briefing lives at **Intelligence** → **Daily Briefing** under Preferences.

What happens when I upload an invoice?

Muin's AI reads it automatically — vendor details, invoice number, dates, line items, tax, total, and payment terms — and files them as structured data with a **confidence score**. No button to press; extraction is part of document processing. Review the result on the invoice detail page: AI-extracted values are marked, and you can edit any field to correct them. Duplicates (same vendor + invoice number) are flagged before they become double payments.

The AI extracted a wrong value from my invoice — what do I do?

Edit the field on the invoice detail page — your correction takes precedence over the AI's reading everywhere downstream (approvals, reports, exports). Start with low-confidence fields: the confidence score tells you where to look first. Totals, dates, and identifiers deserve a glance even at high confidence, since they drive payments.

What is Business Memory and where do its entries come from?

Business Memory is the institutional knowledge Muin learns about your organization — facts, decisions, outcomes, patterns, and relationships — extracted automatically from chats, documents, workflows, and agent runs, or added manually with **Add Memory**. The AI draws on this memory for context, which is how it can answer questions like "what are Acme's payment terms?" with *your* organization's actual answer. Owners and admins manage it at **Intelligence** → **Memory**.

Muin remembered something wrong — how do I fix it?

Open **Intelligence** → **Memory**, find the entry (filters and search help), and mark it **Disputed** — disputed memories are held back from AI context. If the fact changed rather than being wrong, also click **Add Memory** to record the current version. Entries that should never have been captured can be deleted outright. A quick weekly pass over **Auto-extracted** entries keeps memory trustworthy.

What does verifying a memory actually do?

Verified memories carry more weight when the AI assembles context for answers, agents, and sidebars — and disputed ones are silenced. Verification records that a person confirmed the entry, so downstream AI behavior rests on knowledge someone vouched for. Auto-extracted entries are usually right, but the consequential ones (money, terms, deadlines, commitments) deserve a quick check against reality before you rely on them.

How do I create a reminder by just typing what I want?

On the **Reminders** page, use the **AI Reminder** box: type something like *"Remind me to call the Akram Foundation in 3 days at 2pm"* and Muin parses the timing and title for you. The same works conversationally in Chat. If your phrasing can't be parsed, the quick-add form right below takes a title, date, and time — and **More options** opens recurrence, priority, and record linking.

Can reminders repeat automatically?

Yes — open **More options** when creating (or edit an existing reminder) and set the recurrence: daily, weekly, or monthly. Completing one occurrence automatically schedules the next. Recurring reminders fit rhythms like payroll submission or monthly grant-report checks: set once, and the next nudge always exists.

What happens when I snooze a reminder?

Snoozing reschedules the nudge to a later moment — presets like an hour from now or tomorrow morning — without marking the work done. The reminder shows in the **Snoozed** count until it fires again. Snooze is for "not now"; **Complete** is for "done". The summary cards at the top of the Reminders page keep the difference visible.