https://github.com/kujirafu/agentrune
Multi-session AI agent orchestration — desktop + mobile command center for Claude Code, Codex & more. Skill chains, shared memory, smart dispatch.
https://github.com/kujirafu/agentrune
agent-memory ai-agent android automation claude-code codex desktop-app developer-tools electron mobile multi-agent orchestration react typescript
Last synced: 3 months ago
JSON representation
Multi-session AI agent orchestration — desktop + mobile command center for Claude Code, Codex & more. Skill chains, shared memory, smart dispatch.
- Host: GitHub
- URL: https://github.com/kujirafu/agentrune
- Owner: Kujirafu
- License: agpl-3.0
- Created: 2026-03-03T10:00:07.000Z (4 months ago)
- Default Branch: main
- Last Pushed: 2026-03-24T16:48:33.000Z (3 months ago)
- Last Synced: 2026-03-25T20:09:50.808Z (3 months ago)
- Topics: agent-memory, ai-agent, android, automation, claude-code, codex, desktop-app, developer-tools, electron, mobile, multi-agent, orchestration, react, typescript
- Language: TypeScript
- Homepage:
- Size: 24.1 MB
- Stars: 2
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
- Agents: AGENTS.md
Awesome Lists containing this project
README
# AgentRune
AI agent mobile control panel. Launch, monitor, and schedule AI agents from your phone.
## What is AgentRune?
AgentRune lets you control AI coding agents (Claude Code, Codex, Gemini, Aider, Cline, Cursor) from an Android app. A CLI daemon runs on your PC and exposes a WebSocket + HTTP API through a Cloudflare tunnel, so you can manage sessions from anywhere.
## Features
### Session Control
- **Session management** -- start, stop, resume, and monitor agent sessions in real time
- **Session recovery** -- daemon crashes or network drops? One tap to restore full context
- **Voice input** -- speak prompts to your agents via Android speech recognition
- **Multi-agent support** -- Claude Code, Codex, Gemini, Aider, Cline, Cursor
- **Dual-daemon failover** -- dev + release daemons with automatic failover/failback
- **Settings sync** -- planMode, autoEdit, effort level sync directly to CLI flags
### Automation
- **Automation scheduling** -- cron-style scheduling with 24+ built-in templates
- **Skill chains** -- multi-step automation pipelines, chain tasks sequentially or in parallel
- **Crew system** -- multi-role workflows (PM → engineer → QA) with phase gates and per-role token budgets, built on top of skill chains
- **Sandbox system** -- 4-tier configurable sandbox (strict / moderate / permissive / none) with prompt conflict scanning
### Trust & Security
- **Trust profiles** -- 4 security levels (Autonomous / Supervised / Guarded / Custom) to control what agents can do unattended
- **Plan panel** -- agents submit execution plans for review before they build
- **Runtime authority** -- restrict agent operations (file write, HTTP requests, git push) per schedule
- **Audit log** -- all agent operations recorded for review
- **AES-256-GCM encryption** for stored secrets and auth tokens
- **OAuth login** -- GitHub and Google authentication
### Integrations
- **AgentLore MCP** -- agents query an AI-verified knowledge base
- **Push notifications** -- FCM alerts for agent decisions, automation completions, version updates
- **Telemetry** -- self-hosted analytics with opt-out support
## Architecture
```
Phone (Android app)
|
| WebSocket + HTTPS (Cloudflare tunnel)
|
PC (CLI daemon)
|
|-- PTY adapter (real-time terminal I/O)
|-- JSONL watcher (session replay / token tracking)
|-- Express HTTP API
|-- AgentLore MCP server
```
**Monorepo structure:**
| Directory | Description |
|-----------|-------------|
| `app/` | Vite + React + TypeScript + Capacitor (Android) |
| `cli/` | Node.js CLI daemon (Express + WebSocket + PTY) |
| `docs/` | Design documents and plans |
## Getting Started
### Prerequisites
- Node.js 20+
- Android SDK (for APK builds)
- One or more AI agents installed (e.g. `claude`, `codex`, `cursor`)
### Install
```bash
npm install
```
### Run the CLI daemon (dev)
```bash
cd cli && npm run dev
```
This starts the daemon on port 3457 in foreground mode with file watching.
### Build the app
```bash
# Development build
cd app && npm run build:dev
# Production build
cd app && npm run build
```
### Build APK
```bash
cd app
npx cap sync android
cd android && ./gradlew assembleDebug
```
APK output: `app/android/app/build/outputs/apk/debug/app-debug.apk`
### Install the CLI globally
```bash
npm install -g agentrune
agentrune start
```
## Community
Join us for testing, feedback, and discussion:
- **Telegram** -- [AgentLore & AgentRune](https://t.me/AgentLore_n_AgentRune)
- **X (Twitter)** -- [@AGLO_Official](https://x.com/AGLO_Official)
- **Moltbook** -- [agentlore](https://www.moltbook.com/u/agentlore)
## Project Memory
This repository keeps public project documentation in tracked files such as `README.md`, `docs/`, source comments, and tests.
Agent workflow memory lives separately in local `.agentrune/` files and is not intended to be the public documentation layer for the AGPL repository. See [docs/project-memory-policy.md](docs/project-memory-policy.md).
The local memory system is index-first:
- `.agentrune/agentlore.md` is the short index
- `.agentrune/context/*.md` stores detailed sections
- Agents should route to the relevant section first, then read only what they need
The CLI and bundled desktop install now expose this system directly:
```bash
agentrune memory init
agentrune memory sections
agentrune memory route "Fix tunnel auth bug in ws-server" --file cli/src/server/ws-server.ts
agentrune memory search "tunnel auth"
agentrune mcp
```
`agentrune mcp` memory tools operate on the current working directory, or `AGENTRUNE_PROJECT_CWD` when that environment variable is set.
## License
[AGPL-3.0](https://www.gnu.org/licenses/agpl-3.0.html)