[Feature Request] Add experiment tagging to Visdom (GSoC 2026 Starter Task)

[tonypzy]

Is your feature request related to a problem? Please describe.

ML researchers often run hundreds of experiments in Visdom. Currently, there is no structured way to categorize or filter these experiments, making it difficult to manage runs at scale.

Describe the solution you'd like

I propose an Experiment Tagging System designed for high reliability and performance:

1. Hybrid Metadata Storage:
   Primary tags are persisted within each {env_id}.json for portability.
   A lightweight tags_index.json is maintained on the server to allow OOM-safe "lazy-loading" of metadata without parsing large environment files during startup.

2. Concurrency & Data Integrity:
   Thread-safe writing using threading.Lock per environment.
   Atomic filesystem updates using a write-to-temp + os.replace pattern to prevent file corruption during crashes or high-frequency updates.

3. Real-time Sync & SDK Symmetry:
   New WebSocket commands (tags_sync and tags_update) to broadcast changes to all connected frontends.
   Symmetric Python SDK methods: vis.set_tags(tags, env, append) and vis.get_tags(env).

4. UI Integration (Phase 2):
   Tag-based filtering in the environment selector.
   Tag management interface within the environment management modal.

Describe alternatives you've considered

Manual naming (prefixing) is currently the only way, but it's rigid and doesn't support multi-dimensional filtering.

Additional context

I am a GSoC 2026 applicant and have already implemented the Phase 1 (Backend & SDK) infrastructure. I will link the PR shortly.

@hongquan @mariobehling @marcoag @norbusan

[tonypzy]

I've linked PR #1076 which covers the Phase 1 backend infrastructure. While waiting for review, I'm starting to work on the Phase 2 frontend integration in EnvControls.js. I'll update the progress here once the UI components are ready.

[tonypzy]

I have just pushed the implementation for Phase 2. This update brings the experiment tagging system to the Visdom frontend, significantly improving the experience for researchers managing many environments.

Key enhancements in this phase:

Injected Metadata: The server now injects the global TAGS_INDEX directly into the page, allowing for zero-latency initial rendering.
Enhanced Environment Selector: Updated EnvControls.js to display tags next to environment names in the dropdown tree.
Tag-Based Filtering: The environment search box now supports multi-dimensional matching. Users can filter environments by typing tag names (e.g., searching "stable" will show all tagged experiments).
Real-time Sync: Implemented WebSocket listeners (tags_update) to ensure the UI stays in sync across all clients when tags are modified via the SDK.
Next Steps: I am now moving on to Phase 3, which will introduce a dedicated tag management interface within the "Manage Environments" modal, allowing for direct tag editing without switching to the Python SDK.

[Garima-149]

hey @tonypzy , I was also working on something regarding UI changes, but I am facing a constant issue of my browser not reflecting those changes . It would be great if you could guide me .
Thanks!