Auwai

Query

A query resolves a question to chunks and reports the coverage state. It does not read the chunk body.

Flow

The query is a pointer step. The read is the evidence step.

./auwai query "session token" --repo /path/to/repo
./auwai query "session token" --repo /path/to/repo --json

The text output starts with a partition summary:

broker: 3 result(s), 42 unread, 1 stale
app: 1 result(s), 18 unread, 0 stale

Then each matching chunk is printed with its path, line range, token count, and read state.

Coverage

./auwai coverage --repo /path/to/repo
./auwai coverage --repo /path/to/repo --json

The totals are first-party chunks, read chunks, chunks with findings, stale rows, and tokens.

Read the result

./auwai read ID --repo /path/to/repo

The query returns a pointer. The read returns the body and spends the session budget.

For a coverage-first answer, check coverage before the query. A partition with many unread chunks is a gap, not a negative search result.

Current limits

The index is lexical. A concept that is absent from the path, symbol, purpose, and body digest can be missed. The query does not use embeddings. The partition summary is a top-level path, not a semantic category.