Compatibility and data format

Database compatibility

The exporter reads session, message, and project tables from OpenCode’s local SQLite database. It needs session IDs, project IDs, directories and titles; message IDs, session IDs and JSON data containing assistant-message timestamps; and project IDs and names. The exact schema can change between OpenCode releases.

The extractor opens the source read-only, creates a consistent SQLite backup, and validates required columns before reading. If validation fails, it stops with the missing table or columns listed in the error.

Use --db to provide the database path explicitly. Otherwise, the extractor checks OPENCODE_DATA_DIR, XDG_DATA_HOME, and the default per-user OpenCode data directory for opencode.db. Verify the detected path against your installed OpenCode configuration if you have customised its storage location.

CSV format

The CSV includes a source column and one row per session, date, provider, model, and variant. It records input, output, cache-read, cache-write, and reasoning token counters separately. total_tokens is input plus output. cost_usd is the recorded OpenCode cost for the row.

session_title is omitted unless --include-session-title is passed. Session IDs are retained to support identifying repeated records when reviewing exports. Do not share exports without checking for sensitive information.

Source schema

The extractor relies on the message.data JSON object containing assistant message fields (role, time.created, providerID, modelID, variant, cost, and tokens) and on session/project metadata for project identification. This reflects the current OpenCode implementation and may change without notice.