nice work on this. local-first is the right call -- ive seen too many "AI terminal" tools phone home constantly for no good reason.
one thing worth exploring: multi-agent session management gets hairy fast. once you're running 3+ concurrent claude instances, the real pain isnt launching them, its knowing which session did waht and picking it back up later. full-text search across session history turns out to be the killer feature nobody asks for until they desperately need it.
Thanks! Really appreciate the kind words on the local-first approach.
On the multi-agent front, we actually already have run naming, per-agent filtering, and session resume/fork, so identifying and picking up sessions is pretty well covered today. Search currently works across prompts and run metadata, but we're planning to extend it to full conversation content (tool calls, responses, etc.) in a future release. That's the missing piece for the "what happened where" problem you described.
one thing worth exploring: multi-agent session management gets hairy fast. once you're running 3+ concurrent claude instances, the real pain isnt launching them, its knowing which session did waht and picking it back up later. full-text search across session history turns out to be the killer feature nobody asks for until they desperately need it.
On the multi-agent front, we actually already have run naming, per-agent filtering, and session resume/fork, so identifying and picking up sessions is pretty well covered today. Search currently works across prompts and run metadata, but we're planning to extend it to full conversation content (tool calls, responses, etc.) in a future release. That's the missing piece for the "what happened where" problem you described.
Great feedback, it's exactly where we're headed.