Managing Knowledge
Quarterback builds a knowledge base for your org over time. Some knowledge is extracted automatically from conversations (see Turn Extraction), and you can also create items explicitly.
Searching Knowledge
Athena searches your knowledge base automatically before answering technical questions. You can also ask directly:
- “What’s our approach to error handling?”
- “Do we have a pattern for database migrations?”
- “What was the fix for the Redis connection issue?”
Under the hood, this uses the cogz_fetch tool which runs a semantic search across your org’s knowledge. Write your questions naturally — semantic search works better with descriptive queries than keywords.
Filtering by Category
You can narrow searches by asking for specific types:
- “Find any rules about testing”
- “What decisions have we made about authentication?”
- “Are there error patterns for deployment failures?”
Available categories include: rules, decisions, codebase insights, error patterns, preferences, atoms (reusable workflows), pipelines, people, tooling, goals, and more.
Filtering by Date
You can scope searches to a time range:
- “What did we work on last week?”
- “Find decisions made after January 2025”
Creating Knowledge
Tell Athena to remember something, and it will store it:
- “Remember that we always use pnpm, never npm”
- “Save this as a decision: we chose Zod over Joi for validation because of TypeScript inference”
- “Record this error pattern: ECONNREFUSED on port 5432 means Postgres isn’t running, fix with
just dev-infra”
Athena picks the right knowledge type based on what you’re saving and writes it to your org’s knowledge base.
What You Can Store
| Type | When to Use | Example |
|---|---|---|
| Rules | Standards that always apply | ”Always use snake_case for DB columns” |
| Decisions | Settled architectural choices | ”We chose PostgreSQL over MongoDB” |
| Codebase insights | Patterns, conventions, gotchas | ”The gateway caches L3 context for 1 hour” |
| Preferences | How the team likes to work | ”Use Vitest, not Jest” |
| Error patterns | Known errors with solutions | ”SIGTERM on deploy → increase health check timeout” |
| Atoms | Reusable units of work with steps | ”Deploy to staging: step 1, step 2, step 3…” |
| Pipelines | Multi-step workflows orchestrating atoms | ”Release workflow: build → test → deploy → verify” |
| People | Team members and expertise | ”Alice — DevOps lead, expert in Terraform and K8s” |
| Tooling | Systems of record | ”Pinecone — vector DB for semantic search” |
| Goals | Team objectives | ”Ship billing v2 by end of Q1” |
Deduplication
When you save something similar to an existing item, the platform detects the overlap and either updates the existing item or skips the duplicate. You’ll see a message if this happens.
To explicitly replace an older item, tell Athena: “Update the existing rule about testing to also require integration tests.”
Knowledge Scopes
Items are scoped to where they’re most useful:
| Scope | Who Sees It | Examples |
|---|---|---|
Platform (qb) | Everyone, read-only | Platform-shipped baselines |
| Org | All org members | Team rules, decisions, patterns |
| Individual | Only you | Personal preferences, shortcuts |
Most knowledge is stored at the org level. Platform-level items are baselines shipped with Quarterback — you can’t modify them directly, but you can create org-level versions that override them.
How Knowledge Is Used
Knowledge isn’t just stored — it’s actively used:
- Rules and decisions (L3) are injected into every conversation automatically. Athena always follows them.
- Recent turns (L2) from your last ~15 turns are auto-injected so Athena has recent context.
- Everything else is searchable on-demand via
cogz_fetchwhen Athena needs deeper context.
This means the more your team uses Quarterback, the smarter it gets about your codebase, conventions, and preferences.