Different topic from #274, but found while trying to answer a simple question for myself: "what's actually running right now, across my whole project?"
What I expected to find: some kind of live-sessions view — every agent conversation, its status, which task it's tied to, when it started.
What's actually there: the backend fully supports this. GET /api/v1/projects/{id}/conversations returns exactly that — task, agent, status (finished/failed/etc), timestamps, all of it, project-wide. But there's no UI list view anywhere for it. The only frontend route is the per-conversation detail page (/projects/$projectId/conversations/$conversationId), and the only way to actually reach one is by opening a specific task, scrolling its activity feed, and clicking the "started an AI session" link on that one entry — assuming that particular activity happens to have a conversation ID attached to it.
So in practice: no project-wide visibility, no way to browse, no way to tell what's actually running (or silently died) without already knowing which task to check and having a conversation link survive in its activity history.
Why this actually matters for us specifically: we lean on an autonomous agent picking up a fair number of tickets, and it's genuinely hard to answer "is anything actually running right now" at a glance. I ended up writing my own audit tooling against the Postgres-backed session data just to get visibility that — it turns out — the API already exposes cleanly. A UI would've saved that entirely.
What we'd imagine, loosely — you clearly already have the right data model, this really does look like "just" a frontend page:
- A project-level Conversations page (nav item, not just a deep link) listing conversations with status, task title/link, agent name, started time
- Filterable by status (running vs finished vs failed) since "what's still going" and "what silently failed" are the two things you actually want to see fast
- Link each row into the existing detail view — that part's already built
Not sure if this is a "hasn't been prioritized yet" thing or a "there's a reason it's scoped this way" thing — genuinely just flagging it as a real gap we hit, not assuming there's no rationale for the current shape. Happy to elaborate on our use case if it helps.
Different topic from #274, but found while trying to answer a simple question for myself: "what's actually running right now, across my whole project?"
What I expected to find: some kind of live-sessions view — every agent conversation, its status, which task it's tied to, when it started.
What's actually there: the backend fully supports this.
GET /api/v1/projects/{id}/conversationsreturns exactly that — task, agent, status (finished/failed/etc), timestamps, all of it, project-wide. But there's no UI list view anywhere for it. The only frontend route is the per-conversation detail page (/projects/$projectId/conversations/$conversationId), and the only way to actually reach one is by opening a specific task, scrolling its activity feed, and clicking the "started an AI session" link on that one entry — assuming that particular activity happens to have a conversation ID attached to it.So in practice: no project-wide visibility, no way to browse, no way to tell what's actually running (or silently died) without already knowing which task to check and having a conversation link survive in its activity history.
Why this actually matters for us specifically: we lean on an autonomous agent picking up a fair number of tickets, and it's genuinely hard to answer "is anything actually running right now" at a glance. I ended up writing my own audit tooling against the Postgres-backed session data just to get visibility that — it turns out — the API already exposes cleanly. A UI would've saved that entirely.
What we'd imagine, loosely — you clearly already have the right data model, this really does look like "just" a frontend page:
Not sure if this is a "hasn't been prioritized yet" thing or a "there's a reason it's scoped this way" thing — genuinely just flagging it as a real gap we hit, not assuming there's no rationale for the current shape. Happy to elaborate on our use case if it helps.