The problem
A 12-person team — marketing, sales, operations. Documentation existed, but it was scattered across Notion workspaces, Google Docs, and random PDFs on a shared drive. Every time someone asked "how do we process inbound leads?" or "where's the SOP for invoicing?", a 15-minute treasure hunt began.
New hires spent their first two weeks just figuring out where things live and how things work. Team leads got pulled into the same basic questions over and over. The docs were written — nobody could find them fast enough.
What I built
Set up OpenClaw as the team's internal knowledge assistant with full access to their documentation:
1. Structured knowledge base
Consolidated all SOPs, playbooks, checklists and process docs into a structured folder on the server — Markdown and PDF files organized by department. The agent reads these via File System Skill and can search across all of them in seconds.
2. Telegram bot in the team chat
Added the bot to the company's internal Telegram group. Anyone can ask a question — the agent finds the answer in the knowledge base and replies with a direct reference to the source document. No guessing where the info came from.
3. Real questions it handles daily
- "How do we handle inbound leads?" — returns the step-by-step SOP from leads-processing.md
- "Where's the invoicing process?" — finds the file and quotes the key steps
- "What's the campaign launch checklist?" — serves the checklist from campaign-launch.md
- "Who reviews creatives before publishing?" — answers from team-roles.md
- "How do I set up VPN for remote work?" — walks through the setup guide
4. Honest when it doesn't know
If the question isn't covered in the docs, the agent says so explicitly: "I couldn't find this in the documentation. Try asking [responsible person] or check #ops-questions channel." No hallucinated answers.
Result
Unexpected side effect: the team started documenting more proactively, knowing that anything they add to the knowledge base becomes instantly searchable through the bot.
Stack
- VPS: Hetzner CX11 (Ubuntu 22.04, 1 vCPU / 1 GB RAM) — ~$4/mo
- OpenClaw + Telegram Gateway
- ChatGPT (GPT-4o) via oAuth
- File System Skill for knowledge base access
- ~30 Markdown files + PDF documents
Setup time
8 hours — most of it went into gathering and restructuring scattered documentation from different sources. The actual OpenClaw setup took 2 hours. Testing with the team — another hour.
Docs exist but nobody can find them?
I'll set up an AI assistant that knows all your processes and answers your team in seconds.
Message on Telegram