[Feature Request] Support historical session ingestion from OpenCode SQLite databases

[kevincojean]

Hello, and thank you for the great plugin.

Currently, opencode-mempalace tracks and mines session history for new conversations.
But there is no way to retroactively ingest previous sessions' historical data and conversations into the Palace.

I think we'd need the plugin to be able to connect to the OpenCode SQLite session database, extracts historical conversations, and process all of them into the Knowledge Graph and Vector DB. Similar to how mempalace does it for file-based based transcripts (like Claude Code/Codex JSONL files).

What do you think ?

Cheers,

[Necmttn]

I would keep this as a bounded historical backfill path, separate from the live hook mining that currently runs through chatMessage / idle events.

Concrete contract:

• discover the OpenCode SQLite DB and schema version first, then run a read-only snapshot/copy so WAL state or a locked app cannot corrupt the import;
• map sessions, messages, and parts into the same "conversation text for mining" shape the plugin already sends to MemPalace;
• use stable source ids such as DB path hash + table + row id, so rerunning the backfill is idempotent;
• write per-row warnings for malformed or unsupported parts, but keep importing the rest of the session;
• expose --dry-run, --since, and --limit so a user can test a large history safely.

The regression fixture I would want is two historical sessions, one with tool/result parts, one with a malformed row, then a second import proving it does not duplicate memories.

---

Generated with ax.

[kevincojean]

I have created a project for this purpose https://github.com/kevincojean/opencode-mempalacebackfill