The doc chat that ends the policy-lookup pings

• The procedures exist. They are clear, well-written, up to date. They are also unread.
• Onboarding is one senior person answering the same five Slack messages twice a week.
• Anything longer than three pages turns into a just-ping-me-I'll-find-it response.

Searching beats reading. You build a chat over the docs, locked to your domain.

1. 01
   Studio → Agents → Create agent → name it after the team.

   An empty agent shell ready for resources, knowledge, and access rules.

2. 02
   Upload procedures as Resources (PDF, Markdown, exported Notion pages).

   Each one becomes a resource the assistant can read directly.

3. 03
   Click Build knowledge graph. Prostir parses the docs into linked nodes.

   Topics, procedures, owners, and definitions all get connected. Search runs over the graph, not raw text.

4. 04
   Turn on the knowledge search tool.

   The published agent now exposes one search tool that returns the most relevant nodes plus the source file name and section heading.

5. 05
   Set access: only users on @yourcompany.com.

   Outside ChatGPT accounts that try to add the agent get blocked. Internal accounts sign in once.

6. 06
   Publish.

   Link: prostir.build/yourcompany/sops. Drop it into Slack with one line: add this to your ChatGPT.

Whatever you already have. No rewriting. The graph does the linking.

Plain question, cited answer.

• The senior person stops being a search engine. They go back to actual ops work.
• Updates to a doc flow into the next answer. No more stale screenshots in Slack threads.
• When the company hires a new team, you reuse the same agent. Add their docs as Resources.