VibeCockpit

The cockpit for your copilots.
Browse, search, and resume AI coding sessions across all your tools — from one dashboard.

Why?

You vibe-code across multiple AI tools. Claude Code, Claude Desktop, Codex, Copilot, Gemini, OpenCode, Cursor, Antigravity — each stores sessions in its own format and directory. Switching between them means losing track of where you left off.

VibeCockpit scans all of them automatically and gives you one place to search, filter, and resume any session — locally or on remote machines via SSH.

Quick Start

# Install (Linux / macOS)
curl -fsSL https://raw.githubusercontent.com/NJannasch/vibecockpit/main/install.sh | bash

# Launch the web dashboard
vibecockpit --web

# Or use the terminal TUI
vibecockpit

That's it. VibeCockpit auto-detects which tools you have installed and shows all your sessions.

Supported Tools

Tool	Auto-detect	Resume	Remote SSH
Claude Code	✅	✅	✅
Claude Desktop (Claude Code in the desktop app)	✅ (macOS)	✅	—
Codex CLI	✅	✅	✅
GitHub Copilot CLI	✅	✅	✅
OpenCode	✅	—	✅
Gemini CLI	✅	—	✅
Cursor Agent (CLI + IDE)	✅	✅	—
Antigravity (Google)	✅	✅	—
Custom agents (Hermes, etc.)	via SSH	—	✅

Zero config required. If the tool's data directory exists (~/.claude, ~/.codex, etc.), VibeCockpit finds it.

Features

Dashboard — overview of all tools with session counts, active indicators, and recent activity
Cross-tool memory — full-text search across the content of every transcript across every tool (FTS5-backed), exposed via the web UI, the vibecockpit memory search CLI, and the search_memory MCP tool. Any agent connected via MCP can recall what you (or a past agent) said in any prior conversation — Claude Code → Cursor → Codex all share the same memory substrate. Index file (~/.config/vibecockpit/cache/memory.db) can be exported and imported to ship your memory to another machine.
Cost tracking — see what your AI coding tools cost you across all providers. Heatmap, trend charts, per-project breakdown, and plan recommendations that tell you which subscription tier actually fits your usage
Fuzzy search — type hermes or model:opus branch:main active to find anything
Filter chips — click to filter by tool, model, or active status
Group by tool, project, or date — collapsible sections
File tree view — see all projects as a filesystem tree, tool-agnostic
Model picker — resume with a different model (including 1M context variants)
Remote SSH scanning — scan sessions on remote machines, resume via SSH terminal
Dark / light theme — persisted to config
Plugin architecture — extensible for custom tools and agents
Terminal TUI — keyboard-driven alternative with the same search and filtering

Install

One-line install (Linux / macOS)

curl -fsSL https://raw.githubusercontent.com/NJannasch/vibecockpit/main/install.sh | bash

To update, run the same command — the installer detects your existing version and upgrades in place.

macOS note: If you download a binary manually (not via the install script), macOS may show a Gatekeeper warning. Fix with: xattr -d com.apple.quarantine ~/.local/bin/vibecockpit
The install script handles this automatically.

More install options

# With wget
wget -qO- https://raw.githubusercontent.com/NJannasch/vibecockpit/main/install.sh | bash

# Specific version
VERSION=v0.1.0 curl -fsSL https://raw.githubusercontent.com/NJannasch/vibecockpit/main/install.sh | bash

# Custom directory
PREFIX=/opt curl -fsSL https://raw.githubusercontent.com/NJannasch/vibecockpit/main/install.sh | bash

# From source (Docker — recommended; produces a host-platform binary)
git clone https://github.com/NJannasch/vibecockpit && cd vibecockpit
./build.sh                   # Docker handles Node + Go inside the container

# Or build with your own toolchain (Node 22+ and Go 1.24+ required)
./build.sh local

# Download binary directly
# → https://github.com/NJannasch/vibecockpit/releases

Start on login (optional)

vibecockpit --autostart       # systemd (Linux) or launchd (macOS)
vibecockpit --remove-autostart

Configuration

Config is at ~/.config/vibecockpit/config.yaml. Most settings can also be changed from the web UI settings page.

terminal: default                 # auto-detected, or: kitty, alacritty, wezterm, ghostty, ...
theme: light                      # light | dark
sort_by: modified                 # modified | created | name | messages
group_by: none                    # none | provider | project | date

# Override binary paths (e.g. tools installed via nvm)
provider_paths:
  gemini: ~/.nvm/versions/node/v22.21.1/bin/gemini
  opencode: ~/.nvm/versions/node/v22.21.1/bin/opencode

# Scan remote machines via SSH
remote_sources:
  - name: my-server
    host: dev.example.com
    user: deploy
    method: ssh     # or http (for remote VibeCockpit instances)

Tip: When a tool can't be found in PATH, VibeCockpit shows the exact config to add — including an "Open Settings" button in the web UI.

TUI Keyboard Shortcuts

Key	Action
`↑` `↓` `j` `k`	Navigate
`Enter`	Resume session
`m`	Pick model, then resume
`/`	Fuzzy search + filters
`s`	Cycle sort
`Tab`	Cycle grouping
`n`	New project
`d`	Delete session
`q`	Quit

How It Works

VibeCockpit reads session data directly from each tool's local storage. It never modifies your session files (except when you explicitly delete a session).

Tool	Data Location
Claude Code	`~/.claude/projects//sessions-index.json` + `.jsonl`
Claude Desktop	`~/Library/Application Support/Claude/claude-code-sessions///local_*.json` (macOS)
Codex CLI	`~/.codex/state_5.sqlite` + `history.jsonl`
Copilot CLI	`~/.copilot/session-state/*/workspace.yaml` + `events.jsonl`
OpenCode	`~/.local/share/opencode/opencode.db`
Gemini CLI	`~/.gemini/tmp//chats/.json`
Cursor Agent (CLI)	`~/.cursor/chats/<hash>/<uuid>/store.db`
Cursor Agent (IDE)	`~/.config/Cursor/User/globalStorage/state.vscdb`
Antigravity	`~/.gemini/antigravity/brain/*/task.md.metadata.json`

Resuming a Claude Desktop session uses the claude://resume?session=… deep link, opening the conversation back in the Claude Desktop app. Sessions deduped across the standalone CLI and the desktop wrapper so each appears once.

For remote machines, it runs scan commands over SSH — no agent installation required on the remote.

Cost Tracking

VibeCockpit extracts token usage from your session files and estimates costs using current API pricing. No API keys or billing access needed — it's all computed locally from the data already on your machine.

What you get:

Per-session cost estimates on every session card ($12.40 or ~$3.50 for estimates)
GitHub-style activity heatmap — switch between cost, tokens, messages, or session count
Stacked trend chart — daily, weekly, or monthly, grouped by tool or project
Plan recommendations — compares your actual usage against Claude Max 5x/20x, ChatGPT Pro, Gemini Ultra, and tells you which tier fits. Includes estimated API-equivalent value per plan so you can right-size.
Model mix analysis — flags if you're spending 90% on Opus when Sonnet would work for routine tasks
Week-over-week trends, top projects by cost, unused tool alerts

Supported token sources:

Provider	Token data	Cost accuracy
Claude Code	Input, output, cache read/write per message	Exact
OpenCode	Input, output, reasoning, cache per message	Exact
Gemini CLI	Input, output, cached, thoughts per message	Exact
Codex CLI	Total tokens per session	Estimated (70/30 split)
Copilot CLI	No token data (flat subscription)	N/A
Cursor Agent (CLI)	Estimated from message blob sizes	Estimated
Cursor Agent (IDE)	Estimated from message count	Estimated
Antigravity	Conversations encrypted, no token data	N/A
Remote (SSH)	Not yet supported	—

Remote / Headless Mode

VibeCockpit can run headless on a server (e.g. a home lab or VPS) and act as a central memory hub for multiple laptops.

# On the server — bind to LAN, require a token
export VIBECOCKPIT_TOKEN=$(openssl rand -hex 32)
vibecockpit --web --bind 0.0.0.0 --port 3456 --autostart

# On each laptop — pull memory periodically and ship it to the server
vibecockpit memory export ~/.cache/vc.db
curl -X POST -H "Authorization: Bearer $VIBECOCKPIT_TOKEN" \
  -F "file=@$HOME/.cache/vc.db" \
  http://server.local:3456/api/memory/import

By default the web UI binds to 127.0.0.1 only. --bind 0.0.0.0 (or any non-loopback address) requires --token (or VIBECOCKPIT_TOKEN); VibeCockpit refuses to start without one — there's no other auth on the API.

For sensitive networks, terminate TLS in front of VibeCockpit (Caddy, nginx, Traefik, Tailscale Funnel) rather than exposing plain HTTP. Loopback requests (including SSH-tunneled ones) still bypass the token, so ssh -L 3456:localhost:3456 server keeps working without any client config.

MCP over HTTP

When enable_mcp: true is set in config.yaml, the same MCP toolset that vibecockpit --mcp exposes over stdio is also served at POST /mcp (the Streamable HTTP transport). Agents on remote machines can hit this endpoint to query the server's memory index without each laptop having to maintain its own copy.

Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "vibecockpit-remote": {
      "url": "http://server.local:3456/mcp",
      "headers": { "Authorization": "Bearer YOUR_TOKEN" }
    }
  }
}

Cursor's ~/.cursor/mcp.json:

{
  "mcpServers": {
    "vibecockpit-remote": {
      "url": "http://server.local:3456/mcp",
      "headers": { "Authorization": "Bearer YOUR_TOKEN" }
    }
  }
}

Once configured, search_memory and get_session_context queries from any client hit the central index — every machine sees the same cross-tool memory.

Contributing

VibeCockpit is built with Go (backend + TUI) and Svelte 5 (web UI).

git clone https://github.com/NJannasch/vibecockpit
cd vibecockpit

# Development: start Go API and Svelte dev server separately
vibecockpit --web --port 3456 &   # API backend
cd frontend && npm run dev         # Svelte with hot-reload (proxies API)

# Full build (Docker — recommended)
./build.sh                         # cross-builds inside Docker
./build.sh local                   # if you'd rather use the host toolchain

# Run tests
go test ./...

Adding a new provider

Create internal/provider/yourprovider/yourprovider.go
Implement the provider.Provider interface
Add an Available() bool check
Register in main.go → buildRegistry()

Architecture

frontend/             Svelte 5 + Vite (web UI)
internal/
  plugin/             Plugin system + registry
    remote/           SSH + HTTP remote scanning
  provider/           Session scanning per tool
    claude/           Claude Code (JSONL)
    claudedesktop/    Claude Desktop's bundled Claude Code (macOS, deep link)
    codex/            Codex CLI (SQLite)
    copilot/          Copilot CLI (YAML + events)
    opencode/         OpenCode (SQLite)
    gemini/           Gemini CLI (JSON)
  search/             Fuzzy matching + filter parsing
  tui/                Terminal UI (bubbletea)
  web/                HTTP server, embeds Svelte build
  config/             YAML config management
  launcher/           Terminal-aware process launching
  install/            Binary install, desktop entry, autostart

CLI Reference

vibecockpit [flags]

  --web                Start the web UI (opens in browser)
  --port int           Port for the web UI (default: 3456)
  --bind string        Address to bind the web UI to (default: 127.0.0.1; use 0.0.0.0 for LAN)
  --token string       Bearer token required for non-loopback requests (or set VIBECOCKPIT_TOKEN)
  --list               List all sessions (table format)
  --list --json        List all sessions (JSON format)
  --version            Print version
  --install            Install to ~/.local/bin (+ Linux .desktop or macOS .app)
  --uninstall          Remove the installed binary, app launcher, and autostart
  --autostart          Start on login (systemd/launchd)
  --remove-autostart   Remove autostart service
  --yes                Skip confirmation prompts

vibecockpit memory <subcommand>
  reindex              Walk every provider, write transcript content into the FTS5 index
  search <query>       FTS5 search (porter stemmer; supports prefix, phrase, AND/OR/NOT)
  stats                Document count + index size on disk
  export <path>        Snapshot the index to <path> (e.g. for moving to another machine)
  import <path>        Merge another machine's memory.db; local entries win on conflict

License

MIT — Nils Jannasch

Name		Name	Last commit message	Last commit date
Latest commit History 91 Commits
.github/workflows		.github/workflows
assets		assets
frontend		frontend
internal		internal
.gitignore		.gitignore
.golangci.yml		.golangci.yml
.mcp.json		.mcp.json
CLAUDE.md		CLAUDE.md
Dockerfile		Dockerfile
LICENSE		LICENSE
Makefile		Makefile
README.md		README.md
build.sh		build.sh
go.mod		go.mod
go.sum		go.sum
install.sh		install.sh
main.go		main.go
memory_cli.go		memory_cli.go

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

VibeCockpit

Why?

Quick Start

Supported Tools

Features

Install

One-line install (Linux / macOS)

Start on login (optional)

Configuration

TUI Keyboard Shortcuts

How It Works

Cost Tracking

Remote / Headless Mode

MCP over HTTP

Contributing

Adding a new provider

Architecture

CLI Reference

License

About

Uh oh!

Releases 23

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

VibeCockpit

Why?

Quick Start

Supported Tools

Features

Install

One-line install (Linux / macOS)

Start on login (optional)

Configuration

TUI Keyboard Shortcuts

How It Works

Cost Tracking

Remote / Headless Mode

MCP over HTTP

Contributing

Adding a new provider

Architecture

CLI Reference

License

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases 23

Uh oh!

Contributors

Uh oh!

Languages