Find the work buried in agent transcripts

My work

I designed and built Sift, from transcript ingestion and local indexing to the search interface and the flow for returning to a session.

Local search is separate from optional AI knowledge extraction. Extraction is opt in and uses the configured provider.

Different transcript formats, one search

Claude Code and Codex write different session formats. Sift reads both into a common index while retaining the provider and project context needed to resume the work.

Local retrieval first

Indexing and search use a local SQLite database. Optional knowledge extraction is a separate feature, so reading existing work does not require sending every transcript to an AI service.

Return to the original context

A result is useful when it leads back to the right session and working directory. Sift combines search excerpts with transcript reading and resume actions.

The result

Search, filters and readable session details provide a route back to earlier work. The demonstration uses example transcripts with the real indexing and reading flow.