Features intentionally not built for v0. In rough priority order.
witness share <trace_id> → uploads trace to hosted viewer, returns a public URL.
Closed-source hosted service, free tier with sensible limits.
- Playwright agents (via Agent class detection)
- Claude in Chrome event stream subscription
- Stagehand, Skyvern, Manus (as users request)
witness diff trace_a trace_b — compare two runs of the same task.
Eventually: scheduled runs with alerts on drift.
Aggregate views: spend per day, per model, per task pattern.
Watch a running agent step-by-step as it executes.
OpenLLMetry already instruments these; we just need pricing entries and testing.
Workspaces, shared dashboards, RBAC — cloud only, paid.