What Athena Learns

As you work with Athena, it identifies and stores durable knowledge — things that will be useful in future sessions. Here’s what it captures.

Rules

Constraints that always apply to your org. Rules are injected into every conversation.

• “Use snake_case for database columns”
• “Never import from the internal/ directory in client code”
• “All API responses must include a requestId field”

Decisions

Architectural or technical choices that have been made. Decisions prevent Athena from re-debating settled questions.

• “We chose PostgreSQL over MongoDB for the primary datastore”
• “Authentication uses JWT with Vault Transit signing”
• “The billing system uses a prepaid hours model”

Codebase Insights

Patterns, conventions, and gotchas about your codebase.

• “The useAuth hook must be called inside AuthProvider”
• “Database migrations are in packages/db/migrations/ and numbered sequentially”
• “The gateway caches L3 context for 1 hour”

Preferences

How you and your team prefer to work.

• “Prefer explicit error handling over try/catch blocks”
• “Use Vitest for testing, not Jest”
• “Keep PR descriptions concise — 3 bullets max”

Error Patterns

Known errors and their solutions, so Athena can fix them faster next time.

• “ECONNREFUSED on port 5432 — Postgres container isn’t running, run just dev-infra”
• “Module not found: @platform/logger — Run npm run build in packages/logger first”

Other Types

Athena also tracks atoms (reusable units of work), pipelines (multi-step workflows), people (team members), tooling (systems of record), and more. These are stored on disk and retrieved when relevant.

What Athena Doesn’t Store

• Secrets — API keys, passwords, and tokens are detected and rejected before storage
• Ephemeral context — Temporary debugging steps or one-off questions aren’t stored
• Raw code — Athena stores knowledge about code, not the code itself