3ngram as Your Knowledge Base

Stop building a wiki no one reads. Let your AI search everything at once.

Knowledge bases require maintenance. Someone has to write the docs, keep them updated, and hope people search before asking. 3ngram indexes your existing content and makes it searchable alongside your working memory.

1. Documentation drifts the moment it's written

Your wiki has an architecture page from 6 months ago. It's mostly right but the auth module was rewritten and the deployment process changed. No one updated it. New hires read it and get confused.

3ngram: Index your GitHub repos, Google Docs, and Confluence directly. Content syncs automatically. Search returns the current state of your actual documents alongside decisions and context from your memory layer.

2. Knowledge lives in too many places

The answer to 'how does our auth work?' is split across a Confluence page, three GitHub issues, a Slack thread, and a decision someone made in a Claude session last month. No single search covers all of them.

3ngram: Unified hybrid search across indexed content and memories. One query hits your repos, your docs, and your captured decisions simultaneously. Results ranked by semantic relevance, not which tool they came from.

3. Tribal knowledge stays tribal

The senior engineer who knows why the API is structured that way is on vacation. The decision was made in a conversation 4 months ago. The rationale exists nowhere searchable.

3ngram: Decisions captured during AI sessions persist as searchable memory. Why the API is structured that way, why you chose Postgres over DynamoDB, why the retry logic works the way it does. The reasoning survives the conversation that produced it.

See it in action

Here's what 3ngram returns when you use it as your knowledge base.

>How does our authentication flow work?

Found 3 relevant sources:

From GitHub (backend/docs/auth.md):Auth uses short-lived JWTs (15 min) with refresh tokens stored in HTTP-only cookies. Token rotation on every refresh. Session table tracks active sessions with device fingerprint and IP.

From memory (decision, Feb 3):Switched from session-based auth to JWT after the horizontal scaling incident. Sticky sessions were causing failover issues. Kept server-side session tracking for revocation capability -- pure stateless JWT wasn't acceptable for security.

From Confluence (Security Playbook, page 12):Refresh token rotation follows OWASP guidelines. On suspected token reuse, all sessions for that user are invalidated. Rate limiting: 5 failed attempts per 15 min window, then progressive backoff.

Synthesis:The auth flow is JWT-based with server-side session tracking for revocation. This hybrid approach was a deliberate decision from February to solve scaling issues while keeping security controls. The key architectural constraint: never go fully stateless -- the session table is load-bearing for the security model.

FAQ

What content sources can 3ngram index?

GitHub repositories, Google Docs, and Confluence pages. Content is automatically synced and chunked for semantic search. Local markdown files can also be indexed.

Does 3ngram replace Notion or Confluence?

Not as a document editor. 3ngram indexes your existing documents wherever they live and makes them searchable alongside your AI memory. You keep writing in your preferred tool.

How does search work across memories and content?

Hybrid search combining full-text search and vector similarity (pgvector). Memories and indexed content are searched in parallel. Results include the source document or memory with a relevance score.

Index Everything. Maintain Nothing.

3ngram indexes your existing docs and repos, captures decisions as they happen, and makes everything searchable through your AI. No wiki to maintain.

Request access