Internal Knowledge Agent

Overview

A retrieval-augmented agent that answers employee questions across your wiki, Slack, Google Drive, Notion, GitHub, and ticketing systems, with citations to the source.

Architecture

Code shape

1export const knowledge = agent({
2  model: "claude-sonnet-4",
3  retrieval: permissionAware({
4    sources: [notion, drive, slack, github],
5    filterByUserAcl: (user) => user.permissions,
6  }),
7  answerStyle: "with-citations",
8  abstainOn: "no_authoritative_source",
9});

Illustrative. Actual implementation varies per integration. Code ships in your GitHub org.

Who it's for

• →Companies 50–500 people drowning in 'where do I find...' Slack threads
• →Teams with strong docs but weak discoverability
• →Customer-facing teams that need fast internal answers

Outcomes

• ✓Onboarding time cut 30–50%
• ✓Hours of weekly Slack-question drag eliminated
• ✓Knowledge that scales with the company instead of bottlenecking on senior staff

Capabilities

• ·Connectors for Notion, Google Drive, Slack, Confluence, GitHub, Jira, Zendesk
• ·Permission-aware retrieval that surfaces only what the asker can access
• ·Citation-first answers (every claim links to a source)
• ·Slack and web UI
• ·Usage analytics and content-gap reports

Stack

• ▸Anthropic Claude for synthesis
• ▸Vector store: Turbopuffer or pgvector
• ▸Auth: SSO via Okta / Google / Microsoft
• ▸Permission model honored at retrieval time
• ▸Slack app + lightweight web UI

FAQ