# Build an internal AI assistant from company docs

> Create a simple internal assistant that answers team questions from SOPs, policies, help docs, and product knowledge.

- Canonical: https://gptnavi.com/workflows/internal-ai-assistant-from-company-docs
- Category: Operations
- Difficulty: Advanced
- Setup time: 2-3 hours
- Estimated time saved: 4-10 hours
- Last materially updated: 2026-08-24
- Best for: Small teams, Operations teams, Customer support teams
- Tools: Dify, FastGPT, Feishu Base, n8n

## Quick answer

This workflow turns scattered internal documents into a searchable assistant so teammates can find accurate answers without asking the same questions repeatedly.

## When to use it

- Internal SOP assistant
- Support team knowledge assistant
- Onboarding assistant

## Steps

1. **Audit the knowledge sources** — List the documents, SOPs, FAQs, product notes, and policy pages that should be trusted by the assistant. Tool: Feishu Base. Expected output: A source inventory with owners and freshness notes.
2. **Clean and structure documents** — Remove outdated notes, merge duplicates, add clear titles, and split long pages into focused sections. Tool: Notion. Expected output: A clean document set ready for ingestion.
3. **Create the assistant** — Upload the documents into an AI app builder and configure answer style, refusal rules, and source citation behavior. Tool: Dify. Expected output: A working internal Q&A assistant.
4. **Test with real questions** — Ask the assistant common onboarding, support, and operations questions. Log wrong answers and missing documents. Tool: FastGPT. Expected output: A test report and improvement list.

## Prompt templates

### Knowledge source audit

Review this list of company docs. Identify duplicates, outdated sources, missing SOPs, unclear ownership, and documents that should not be used by an AI assistant. Docs: [paste]

### Assistant system instruction

Write system instructions for an internal company AI assistant. It should answer from approved documents, cite sources when possible, say when it does not know, and avoid inventing policy. Context: [paste company context]

## Common mistakes

- Uploading messy outdated docs
- Not testing with real team questions
- Allowing the assistant to answer without source constraints

## Related workflows

- https://gptnavi.com/workflows/ai-customer-support-knowledge-base
- https://gptnavi.com/workflows/summarize-support-tickets-into-product-insights
- https://gptnavi.com/workflows/meeting-notes-to-action-items