I am running openclaw 2026.4.22 on a Linux vm on a macOS host. The embedder and deriver are local Ollama models running on the host, the reasoning and dreaming are cloud models. The vector database is lancedb instead of pgvector, the rest of the setup is standard, in a docker container. Everything runs fine.
I want to migrate data from an old setup to this machine. I made a file from the raw jsonl conversation files including date, peer, message. I want this imported in the new honcho database. I tried a lot of things but keep running into problems including api limits and refusal of honcho to import messages that are not acquired from the chat/tui/telegram.
I was thinking about the following solution: run an extra agent, and feed it the messages one by one via one of the channels so that honcho thinks they are real. like a replay of conversations. Do you think this could work? Could I trick honcho in accepting dates and times that differ from the real date and time? How should I setup such a feed?