Claude Code web UI for coding, life, and anything in between.
HyperNeo runs a daemon on your machine that drives Claude Code agent sessions inside your workspace and serves a browser UI for them: several sessions side by side, model switching mid-conversation, tasks and goals with approvals, and MCP tools for files, git, and shell work. Sessions and history stay in a local SQLite database.
- Developers who use Claude Code or the Anthropic API and want a browser — including a phone — for agents working in local repositories
- Anyone running several agent sessions at once and tracking their output as tasks and goals
- People routing agents through OpenRouter or Ollama (local or cloud) instead of, or alongside, Anthropic
npm install -g hyperneo
export ANTHROPIC_API_KEY=sk-ant-...
hyperneoOpen http://localhost:9283 and start a conversation.
Already logged in to Claude Code? HyperNeo picks up that login from ~/.claude/.credentials.json or the macOS Keychain, so the export is optional.
Prefer a native app? Releases ship a desktop build: .dmg for macOS (Apple Silicon and Intel), .msi/.exe for Windows x64, and .deb/.rpm for Linux x64. The Linux and Windows installers are unsigned.
hyperneo [options]
| Option | Default | Description |
|---|---|---|
-p, --port <port> |
9283 |
Port to listen on |
--host <host> |
0.0.0.0 |
Host to bind to |
--db-path <path> |
~/.hyperneo/data/daemon.db |
Database file path |
--workspace <path> |
— | Default workspace root for file indexing |
-V, --version |
Show version number | |
-h, --help |
Show help |
Subcommands such as hyperneo send <session-id> <text> and hyperneo build interact with a running daemon; see hyperneo --help. A standalone daemon without the web UI is published as hyperneod.
Set one of these environment variables before starting:
| Variable | Description |
|---|---|
ANTHROPIC_API_KEY |
Anthropic API key (recommended) |
CLAUDE_CODE_OAUTH_TOKEN |
OAuth token from Claude Code |
OPENROUTER_API_KEY |
OpenRouter API key for OpenRouter models |
OLLAMA_API_KEY |
Optional API key for a local Ollama server |
OLLAMA_BASE_URL |
Optional local Ollama URL (defaults to http://localhost:11434) |
OLLAMA_CLOUD_API_KEY |
Ollama Cloud API key |
OLLAMA_CLOUD_BASE_URL |
Optional Ollama Cloud URL (defaults to https://ollama.com) |
OpenRouter models use provider/model IDs such as anthropic/claude-sonnet-4.6 and openrouter/auto. See OpenRouter setup for provider details and compatibility notes.
Ollama local and Ollama Cloud models use the native /api/chat and /api/tags APIs through HyperNeo's provider bridge. See Ollama setup for local and cloud configuration. HyperNeo's bridging also supports GLM, Kimi, MiniMax, DeepSeek, and OpenCode endpoints.
- Multi-session conversations — run multiple independent agent sessions side by side
- Spaces, tasks, and goals — organize work, with approvals before agents act on it
- Model switching — swap models mid-conversation, with extended-thinking budgets from 8k to 64k tokens
- MCP server support — connect external tools via the Model Context Protocol
- Skills — plugin and MCP skills extend sessions; Fetch, Chrome DevTools, and Playwright ship built in
- Rewind — roll back file changes, the conversation, or both, to any earlier point
- Runs on your machine — npm binaries for macOS, Linux, and Windows; desktop installers; and a standalone daemon
This project uses a dev branch for active development. See CONTRIBUTING.md for details on the branching strategy and contribution guidelines.
Quick overview:
- All feature PRs should target the
devbranch - Releases go directly from
devvia version tags - Requires Bun 1.3.8 or later:
bun install, thenbun run devfor a dev server orbun run checkfor the full quality suite
The HyperNeo brand identity materials in docs/branding/hyperneo-visual-identity/ are excluded from Apache-2.0 and are all rights reserved under their asset license. The vector export helper remains Apache-2.0.