[FEATURE] Web UI with multi-project support

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

It would be great to have a web-based UI for Claude Code that allows multiple projects to be open simultaneously in a single interface — similar to what tools like opencode web offer.

When working across multiple repositories or projects at the same time, the current CLI requires separate terminal sessions with no unified view. A web UI would make it easier to context-switch and manage parallel workstreams.

Proposed Solution

What I'd like to see:

• A web interface for Claude Code
• Ability to open and switch between multiple projects in one UI
• Independent agent sessions per project

Usage:

• navigate to my projects dir. Example: /Projects
• run claude web
• open https://127.0.0.1:31212145
• work :)

Alternative Solutions

_No response_

Priority

High - Significant impact on productivity

Feature Category

Other

Use Case Example

I am not an ML Engineer , the following context was generated by mapping my scenarios to a common domain.

Use Case: ML "Model-to-Production" Pipeline

A typical ML release touches at least 4 repos at once:

Session A data-ingestion-pipeline batch → streaming refactor (Spark/Python)
Session B model-training-service DDP boilerplate + experiment hooks (PyTorch)
Session C inference-api ONNX runtime tuning, P99 latency (FastAPI/C++)
Session D ml-infra-as-code GPU autoscaling, Helm config (Kubernetes)

With a Web UI, these all live in one browser tab alongside the tools already running in the browser (experiment dashboards, logging UIs, vector DB portals, cloud consoles).

Concrete workflows this unlocks:

• Copy hyperparameters from an experiment dashboard tab → paste into Session B. No reformatting.
• Grab a raw JSON error payload from a logging UI → drop into Session C to debug a schema validation issue.
• Paste a loss function from a PDF research paper directly into a Claude session — no terminal encoding issues or broken indentation.
• While a GPU cluster spins up in a Cloud Console tab, use another session to write unit tests for the next model version.
• View Session A (feature engineering) and Session C (inference logic) side by side to catch training-serving skew early.

Why the browser specifically

The browser is already where ML engineers spend half their day — experiment trackers, model registries, team planning tools, observability dashboards, cloud consoles. Bringing Claude Code into that environment
removes the constant alt-tab between terminal and browser and makes the clipboard a first-class integration layer.

Minimum viable feature set

• Multiple named project sessions in one UI
• Per-session file tree, diff view, and agent output
• Persistent session state across reloads
• Locally hosted (privacy-first)

Additional Context

Similar features in other tools:

• opencode — web-based coding agent UI with multi-session support. Closest existing reference for what this could look like.

Technical considerations:

• Sessions could map 1:1 to existing claude CLI instances under the hood — the Web UI would essentially be a frontend multiplexer over the existing agent runtime.
• Local hosting is preferred to avoid sending project file trees to a remote server unnecessarily.
• Session isolation is critical — each project should have its own context, working directory, and agent state with no bleed-through.

Related pain points already reported by the community:

• Terminal character encoding breaks when pasting math/LaTeX or large Python blocks.
• No way to visually compare agent output across two repos simultaneously.
• Managing multiple claude sessions in tmux works but has no persistent state or GUI.