Schema and legacy migration¶

All schema migrations live in internal/store/migrations/, are embedded in the Go binary, and retain their original numbers. Opening an archive for writing applies new migrations. Read-only MCP requests never migrate files. SQL table/column names remain compatible with the Python-era archive.

chatstrata import-legacy /path/to/old/chatstrata.duckdb

Legacy import reads the old database's raw events and records observations, then builds the Go projection. It does not need Python or alter the old file. Conversation/message IDs become deterministic Go IDs; labels and embeddings are not migrated. Missing raw events cannot be recovered. OpenCode titles that existed only in a legacy session row may be unavailable.

Keep the old database as a backup. The Go archive remains named chatstrata-go.duckdb and its source log chatstrata-go.log/, preserving existing installations. Back up the log and use chatstrata rebuild to recreate the projection. Do not rename the archive during an upgrade without also moving its corresponding log and updating daemon/MCP configuration.