a5e413c0 · zzmq.dev@gmail.com/zz
Commit
a5e413c0dae78240c1aa262d0a931fc7bd802684
Stream model responses token-by-token instead of buffering Switched the Ollama chat call from stream:false to stream:true and parse the real NDJSON wire format (verified directly: prose content arrives incrementally, tool_calls arrive as one complete chunk once decided -- same total generation time either way, just delivered as it's produced instead of buffered until the end). runAgent gains onPartial(textSoFar), called as content streams in; zz work wires it to the activity tab's status line (tail-truncated to one line, since ActivityLog budgets exactly one row for it) so you see the model's reasoning appear as it's written, the same way Claude Code's own output streams, instead of a silent 10-30s gap per turn. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Troy Howard <thoward37@gmail.com> -- 2026-10-10T19:44:07.000Z