claude-view

You have 10 Claude sessions running right now. What are they doing?

English · 繁體中文 · 简体中文 · 日本語 · Español · Français · Deutsch · Português · Italiano · 한국어 · Nederlands

Behind every "thinking..." spinner, Claude is spawning sub-agents, calling MCP servers, running skills, firing hooks — and you can't see any of it. You Cmd+Tab through 15 terminals trying to remember which session was doing what. A cache expired while you weren't looking. A session finished 10 minutes ago and you didn't notice. Another one hit its context limit and you're burning tokens on a dead conversation.

You're paying $200/mo for Claude Code. You deserve a dashboard.

curl -fsSL https://get.claudeview.ai/install.sh | sh

One command. Every session visible. Real-time.

What You Get

Live Monitor

Feature	Why it matters
Session cards with last message	Instantly remember what each long-running session is working on
Notification sounds	Get pinged when a session finishes or needs your input — stop polling terminals
Context gauge	Real-time context window usage per session — see which ones are in the danger zone
Cache warm countdown	Know exactly when prompt cache expires so you can time your next message to save tokens
Cost tracking	Per-session and aggregate spend — hover for token/cost breakdown with cache savings by category
Sub-agent visualization	See the full agent tree — sub-agents, their status, and what tools they're calling
Multiple views	Grid, List, Kanban, or Monitor mode — pick what fits your workflow
Kanban swimlanes	Group sessions by project or branch — visual swimlane layout for multi-project workflows
Recently closed sessions	Sessions that end appear in a "Recently Closed" section instead of vanishing — persist across server restarts
Unified live chat	History and real-time messages in a single scrollable conversation — no more tab-switching
Queued message indicators	Messages waiting in the queue show as pending bubbles with a "Queued" badge
SSE-driven live data	Live session data pushed via Server-Sent Events — eliminates stale-cache risks entirely

Rich Chat History

Feature	Why it matters
Full conversation browser	Every session, every message, fully rendered with markdown and code blocks
Tool call visualization	See file reads, edits, bash commands, MCP calls, skill invocations — not just text
Compact / verbose toggle	Skim the conversation or drill into every tool call
Thread view	Follow agent conversations with sub-agent hierarchies
Export	Markdown export for context resumption or sharing
Bulk select & archive	Select multiple sessions for batch archiving with persistent filter state
Encrypted sharing	Share any session via E2E encrypted link — zero server trust

Advanced Search

Feature	Why it matters
Full-text search	Search across all sessions — messages, tool calls, file paths
Unified search engine	Tantivy full-text + SQLite pre-filter run in parallel — one endpoint, no frontend wiring gaps
Project & branch filters	Scope to the project you're working on right now
Command palette	`Cmd`+`K` to jump between sessions, switch views, find anything

Plans, Prompts & Teams

Feature	Why it matters
Plan browser	View your `.claude/plans/` directly in session detail — no more hunting through files
Prompt history	Full-text search across all prompts you've sent to Claude Code with template clustering and intent classification
Teams dashboard	See team leads, inbox messages, team tasks, and file changes across all team members — with JSONL fallback so history is never lost
Prompt analytics	Leaderboards of prompt templates, intent distribution, and usage statistics

Plugin Manager

Feature	Why it matters
GUI plugin browser	Browse, install, enable, disable, and uninstall Claude Code plugins without touching a terminal
Plugin card grid	Every installed plugin shown as a card with name, status, scope, and version
Marketplace dialog	Discover new plugins and install with scope picker (user-level or project-level)
Health banner	Plugin system status at a glance — know when something is broken before it breaks your workflow

Workflows

Feature	Why it matters
Workflow builder	Create and run multi-stage workflows from a dedicated page — VS Code-style layout with Mermaid diagram preview and YAML editor
Streaming LLM chat rail	Generate workflow definitions in real time using an embedded chat interface
Stage runner	Visualize stage columns, attempt cards, and progress bar as your workflow executes
Built-in seed workflows	Plan Polisher and Plan Executor ship out of the box

Open in IDE

Feature	Why it matters
One-click file open	Files referenced in sessions have a button to open them directly in your editor
Auto-detects your editor	VS Code, Cursor, Zed, and others auto-detected — no configuration needed
Everywhere it matters	Button appears in Changes tab, file headers, and Kanban project headers
Preference memory	Your preferred editor is remembered across sessions

Agent Internals — See What's Hidden

Claude Code does a lot behind "thinking..." that never shows in your terminal. claude-view exposes all of it.

Feature	Why it matters
Sub-agent conversations	See the full tree of spawned agents, their prompts, and their outputs
MCP server calls	See which MCP tools are being invoked and their results
Skill / hook / plugin tracking	Know which skills fired, which hooks ran, what plugins are active
Hook event recording	Dual-channel hook capture (live + JSONL backfill) — every hook event recorded and browsable, even for past sessions
Worktree branch drift	Detects when git worktree branches diverge — shown in live monitor and history
@file mention chips	`@filename` references extracted from sessions shown as chips on cards — hover for the full path
Agent SDK live chat	Thinking blocks, tool calls, and tool results with syntax highlighting, collapsible sections, and a context window gauge
Tool use timeline	Action log of every tool_use/tool_result pair with timing
Error surfacing	Errors bubble up to the session card — no more buried failures
Raw message inspector	Drill into any message's raw JSON when you need the full picture

Analytics

A rich analytics suite for your Claude Code usage. Think Cursor's dashboard, but deeper.

Dashboard Overview

Feature	Description
Week-over-week metrics	Session count, token usage, cost — compared to your previous period
Activity heatmap	90-day GitHub-style grid showing your daily Claude Code usage intensity
Top skills / commands / MCP tools / agents	Leaderboards of your most-used invocables — click any to search matching sessions
Most active projects	Bar chart of projects ranked by session count
Tool usage breakdown	Total edits, reads, and bash commands across all sessions
Longest sessions	Quick access to your marathon sessions with duration

AI Contributions

Feature	Description
Code output tracking	Lines added/removed, files touched, commit count — across all sessions
Cost ROI metrics	Cost per commit, cost per session, cost per line of AI output — with trend charts
Model comparison	Side-by-side breakdown of output and efficiency by model (Opus, Sonnet, Haiku)
Learning curve	Re-edit rate over time — see yourself getting better at prompting
Branch breakdown	Collapsible per-branch view with session drill-down
Skill effectiveness	Which skills actually improve your output vs which ones don't

Insights (experimental)

Feature	Description
Pattern detection	Behavioral patterns discovered from your session history
Then vs Now benchmarks	Compare your first month to recent usage
Category breakdown	Treemap of what you use Claude for — refactoring, features, debugging, etc.
AI Fluency Score	Single 0-100 number tracking your overall effectiveness

Insights and Fluency Score are in early experimental stage. Treat as directional, not definitive.

Built for the Power User

claude-view is for the developer who:

Runs 3+ projects simultaneously, each with multiple worktrees
Has 10-20 Claude Code sessions open at any time
Needs to context-switch fast without losing track of what's running
Wants to optimize token spend by timing messages around cache windows
Gets frustrated by Cmd+Tab-bing through terminals to check on agents
Worktree-aware — detects branch drift across git worktrees

How It's Built


Blazing fast	Rust backend with SIMD-accelerated JSONL parsing, memory-mapped I/O — indexes thousands of sessions in seconds
Real-time	File watcher + SSE + unified WebSocket with heartbeat, event replay, and crash recovery
Tiny footprint	~10 MB download, ~27 MB on disk. No runtime dependencies, no background daemons
100% local	All data stays on your machine. Zero telemetry, zero required accounts. Optional encrypted sharing available.
Zero config	`npx claude-view` and you're done. No API keys, no setup, no accounts

Why Rust? — The Numbers

Measured on an M-series Mac with 1,493 sessions across 26 projects:

Metric	claude-view	Typical Electron dashboard
Download	~10 MB	150–300 MB
On disk	~27 MB	300–500 MB
Startup (ready to serve)	< 500 ms	3–8 s
RAM (full index loaded)	~50 MB	300–800 MB
Deep-index 1,500 sessions	< 1 s	N/A
Runtime dependencies	0	Node.js + Chromium

Key techniques:

SIMD pre-filter — memchr scans raw bytes before touching a JSON parser
Memory-mapped I/O — JSONL files are mmap'd and parsed in-place, never copied
Tantivy search — Same engine behind Quickwit; indexes 1,500 sessions in under a second
Zero-copy where it counts — Borrowed slices from mmap through parse to response

cargo build --release
/usr/bin/time -l target/release/claude-view   # peak RSS + wall time

How It Compares

The Claude Code ecosystem has great tools — chat UIs, history viewers, session managers. claude-view fills a different gap: real-time monitoring + deep history + analytics in one lightweight workspace.

Tool	Category	Stack	Size	Live monitor	Search	Analytics
claude-view	Monitor + workspace	Rust	~10 MB	Yes	Yes	Yes
opcode	GUI + session manager	Tauri 2	~13 MB	Partial	No	Yes
ccusage	CLI usage tracker	TypeScript	~600 KB	No	No	CLI
CodePilot	Desktop chat UI	Electron	~140 MB	No	No	No
claude-run	History viewer	TypeScript	~500 KB	Partial	Basic	No

Chat UIs (CodePilot, CUI, claude-code-webui) are interfaces for Claude Code. claude-view is a dashboard that watches your existing terminal sessions. They're complementary, not competing.

Why the size difference matters

	claude-view	Electron app
Download	~10 MB	~140 MB
On disk	~27 MB	~400 MB
What's in it	Rust server + SPA assets	Chromium + Node.js + Next.js + app code
RAM at idle	~50 MB	~300 MB+
Startup	< 500 ms	3–8 s
Background cost	Negligible	Chromium renderer process

When you're already running 10+ Claude Code sessions eating RAM and CPU, the last thing you want is a 300 MB dashboard competing for resources.

Installation

Method	Command
Shell (recommended)	`curl -fsSL https://get.claudeview.ai/install.sh \| sh`
npx	`npx claude-view`

The shell installer downloads a pre-built binary (~10 MB), installs to ~/.claude-view/bin, and adds it to your PATH. Then just run claude-view.

Only requirement: Claude Code installed — this creates the session files we monitor.

Configuration

Env Variable	Default	Description
`CLAUDE_VIEW_PORT` or `PORT`	`47892`	Override the default port

Self-hosting / Local Dev — Optional Cloud Features

The pre-built binary (npx claude-view) ships with auth, sharing, and mobile relay baked in. If you're building from source or running your own server, these features are opt-in via environment variables — omit any of them and that feature is simply disabled. No errors, no warnings, just graceful degradation.

Env Variable	Feature	What happens without it
`SUPABASE_URL`	Login / auth	Auth disabled — app runs in fully local, zero-account mode
`RELAY_URL`	Mobile pairing (QR code)	Mobile relay disabled — QR pairing unavailable
`SHARE_WORKER_URL` + `SHARE_VIEWER_URL`	Encrypted session sharing	Share button hidden — sessions stay local

To run fully local with no cloud dependencies (the default for self-hosters):

bun dev    # just works — all cloud features off

To enable a feature, export the env var before starting the server:

export SUPABASE_URL=https://your-project.supabase.co
export RELAY_URL=wss://your-relay.fly.dev/ws
export SHARE_WORKER_URL=https://your-worker.example.com
export SHARE_VIEWER_URL=https://your-viewer.example.com
bun run dev:server

These are runtime overrides — they take precedence over whatever is baked into the binary at build time.

Corporate / Sandbox Environments

If your machine restricts writes to ~/Library/Caches/ (e.g., DataCloak, CrowdStrike, corporate DLP):

cp crates/server/.env.example .env
# Uncomment the CLAUDE_VIEW_DATA_DIR line

This keeps the database, search index, and lock files in .data/ inside the repo — no writes outside the project directory.

Community

Website: claudeview.ai — docs, changelog, blog
Discord: Join the server for support, feature requests, and discussion

Development

Prerequisites: Rust, Bun, cargo install cargo-watch

bun install        # Install all workspace dependencies
bun dev            # Start full-stack dev (Rust + Web with hot reload)

Workspace Layout

Path	Package	Purpose
`apps/web/`	`@claude-view/web`	React SPA (Vite) — main web frontend
`apps/share/`	`@claude-view/share`	Share viewer SPA (Vite) — Cloudflare Pages
`apps/mobile/`	`@claude-view/mobile`	Expo native app
`apps/landing/`	`@claude-view/landing`	Static HTML landing page
`packages/shared/`	`@claude-view/shared`	Shared types & theme tokens
`packages/design-tokens/`	`@claude-view/design-tokens`	Colors, spacing, typography
`packages/mcp/`	`@claude-view/mcp`	MCP server integration
`packages/plugin/`	`@claude-view/plugin`	Claude Code plugin (tools + skills)
`crates/`	—	Rust backend (Axum)
`infra/share-worker/`	—	Cloudflare Worker — share API (R2 + D1)
`infra/install-worker/`	—	Cloudflare Worker — install script proxy with download tracking

Dev Commands

Command	Description
`bun dev`	Full-stack dev — Rust server (cargo-watch) + Web frontend (Vite HMR)
`bun run dev:web`	Web frontend only (assumes Rust server already running)
`bun run dev:server`	Rust backend only (with cargo-watch)
`bun run dev:all`	All JS/TS apps via Turbo (web + mobile + landing, no Rust)
`bun run build`	Build all workspaces
`bun run preview`	Build web + serve via release binary
`bun run lint`	Lint all JS/TS workspaces
`bun run lint:all`	Lint JS/TS + Rust (Clippy)
`bun run typecheck`	TypeScript type checking
`bun run test`	Run all tests (Turbo)
`bun run test:rust`	Run Rust tests (`cargo test --workspace`)
`cd apps/web && bunx vitest run`	Run web frontend tests only
`bun run fmt`	Format Rust code
`bun run cleanupport`	Kill processes on dev ports (47892, 5173)

Testing Production Distribution:

bun run dist:test    # One command: build → pack → install → run

Command	Description
`bun run dist:pack`	Package binary + frontend into tarball at `/tmp/`
`bun run dist:install`	Extract tarball to `~/.cache/claude-view/` (simulates first-run download)
`bun run dist:run`	Run the npx wrapper using the cached binary
`bun run dist:clean`	Remove all dist cache and temp files

Releasing:

bun run release          # patch bump: 0.1.0 → 0.1.1
bun run release:minor    # minor bump: 0.1.0 → 0.2.0
bun run release:major    # major bump: 0.1.0 → 1.0.0

git push origin main --tags    # triggers CI → builds all platforms → auto-publishes to npm

Platform Support

Platform	Status
macOS (Apple Silicon)	Available
macOS (Intel)	Available
Linux (x64)	Planned
Windows (x64)	Planned

claudeview.ai — Official website, docs, and changelog
@claude-view/plugin — Claude Code plugin with 8 MCP tools and 3 skills. claude plugin add @claude-view/plugin
claude-backup — Claude Code deletes your sessions after 30 days. This saves them. npx claude-backup

If claude-view helps you fly Claude Code, consider giving it a star.

tombelieber/claude-view

claude-view

What You Get

Live Monitor

Rich Chat History

Advanced Search

Plans, Prompts & Teams

Plugin Manager

Workflows

Open in IDE

Agent Internals — See What's Hidden

Analytics

Built for the Power User

How It's Built

How It Compares

Installation

Community

Workspace Layout

Dev Commands

Platform Support

On this page

Languages

Contributors

Latest Release

tombelieber/claude-view

claude-view

What You Get

Live Monitor

Rich Chat History

Advanced Search

Plans, Prompts & Teams

Plugin Manager

Workflows

Open in IDE

Agent Internals — See What's Hidden

Analytics

Built for the Power User

How It's Built

How It Compares

Installation

Community

Workspace Layout

Dev Commands

Platform Support

Related

On this page

Languages

Contributors

Latest Release