Summary
AgentV Studio should lean fully into project-first navigation.
I compared Phoenix (https://phoenix.enty.pro/) with AgentV Studio in project mode and identified a set of UX improvements that would make Studio feel more aligned with Phoenix and other eval tools.
Decision on --multi
Decision: hard remove --multi.
Recommended dashboard model:
- default = project-first Studio
- override =
--single
Rationale:
- no known active usage of
--multi
- lower mental overhead for humans and AI agents
- cleaner alignment with the rename from benchmarks to projects
- one stable default mental model is better than dual-mode UX
This means the follow-up should:
- remove the
--multi flag
- remove the
--multi deprecation warning code
- update help text/docs accordingly
- remove remaining user-facing
multi_project_dashboard plumbing that only exists to support --multi
- keep
--single as the only explicit dashboard override
Proposed follow-up changes
1. Remove --multi and simplify dashboard mode behavior
- Make project-first mode the clear default behavior.
- Remove
--multi from the CLI surface.
- Keep
--single as the only explicit dashboard override.
- Ensure docs and help text describe one default mental model.
2. Show project names, not project IDs, on project pages
Current project pages render the slug/ID in the page title instead of the human-facing project name.
Expected:
- Use the project display name prominently in headers and navigation.
- Reserve the ID for internal routing / API identity.
3. Add project list controls on the dashboard
Phoenix makes project discovery much easier with search and view controls.
Add to the AgentV projects dashboard:
- search by project name/path
- sort controls (recent activity, name, pass rate)
- optional grid / table density toggle
- maybe quick filter for “recently active” vs “all”
4. Expose project management actions in the UI
There is API/client support for removing projects, but the projects dashboard does not visibly expose remove/archive management.
Add:
- remove project action
- confirmation UI
- optional archive/hide behavior if we want a softer alternative to delete
5. Improve the Add Project flow
The current flow is a raw filesystem path textbox.
Improve with:
- clearer empty-state copy
- validation/error affordances
- recent paths or suggestions
- optional picker integration if feasible
- better explanation of what qualifies as a project
6. Strengthen project cards
Current cards are functional but thin.
Possible additions:
- experiments count
- targets count
- last run status badge
- trend / freshness signal
- more obvious primary action
7. Make the navigation more consistently project-first
Even in project mode, some navigation still feels run-centric.
Review:
- left sidebar hierarchy and labels
- project-aware breadcrumbs
- consistency between project dashboard, project detail, and run detail pages
Acceptance signals
- Users land in a project-first Studio without needing to think about
--multi.
- The UI consistently says project and presents projects as the top-level object.
- Project discovery and cleanup remain usable once many projects are registered.
- The single-project view remains available via
--single without shaping the default UX.
Relevant code
apps/cli/src/commands/results/serve.ts
apps/studio/src/routes/index.tsx
apps/studio/src/routes/projects/$projectId.tsx
apps/studio/src/components/ProjectCard.tsx
apps/studio/src/components/Sidebar.tsx
apps/studio/src/components/Breadcrumbs.tsx
apps/studio/src/lib/api.ts
Notes
This issue is intended as a handoff artifact from UX comparison work, not as a request for a large redesign. The direction should stay lightweight and pragmatic.
Summary
AgentV Studio should lean fully into project-first navigation.
I compared Phoenix (
https://phoenix.enty.pro/) with AgentV Studio in project mode and identified a set of UX improvements that would make Studio feel more aligned with Phoenix and other eval tools.Decision on
--multiDecision: hard remove
--multi.Recommended dashboard model:
--singleRationale:
--multiThis means the follow-up should:
--multiflag--multideprecation warning codemulti_project_dashboardplumbing that only exists to support--multi--singleas the only explicit dashboard overrideProposed follow-up changes
1. Remove
--multiand simplify dashboard mode behavior--multifrom the CLI surface.--singleas the only explicit dashboard override.2. Show project names, not project IDs, on project pages
Current project pages render the slug/ID in the page title instead of the human-facing project name.
Expected:
3. Add project list controls on the dashboard
Phoenix makes project discovery much easier with search and view controls.
Add to the AgentV projects dashboard:
4. Expose project management actions in the UI
There is API/client support for removing projects, but the projects dashboard does not visibly expose remove/archive management.
Add:
5. Improve the Add Project flow
The current flow is a raw filesystem path textbox.
Improve with:
6. Strengthen project cards
Current cards are functional but thin.
Possible additions:
7. Make the navigation more consistently project-first
Even in project mode, some navigation still feels run-centric.
Review:
Acceptance signals
--multi.--singlewithout shaping the default UX.Relevant code
apps/cli/src/commands/results/serve.tsapps/studio/src/routes/index.tsxapps/studio/src/routes/projects/$projectId.tsxapps/studio/src/components/ProjectCard.tsxapps/studio/src/components/Sidebar.tsxapps/studio/src/components/Breadcrumbs.tsxapps/studio/src/lib/api.tsNotes
This issue is intended as a handoff artifact from UX comparison work, not as a request for a large redesign. The direction should stay lightweight and pragmatic.