v0.3.6

Penglai

P E N G L A I

Bring agents that can work, listen, and speak to the tools you already use

Penglai Desktop
Penglai Agent
πŸ’¬ Chat Workspace
β–Ά Run History
πŸ“‘ Channel Manager
✨ Skill Manager
Today 14:30
Check today's server status
Runtime Hub running normally βœ… CPU 23% Β· Memory 45% 3 alerts resolved
Thanks!
Feishu Channel
Today's server monitoring report
Generating daily report, please wait...
Report ready: CPU 23%, Memory 45%, 3 alerts resolved.
Terminal
$ penglai start
Runtime Hub started :8200
$ penglai chat "Today's summary"
Processing request...
I spent ten years in networking, security, and ops β€” but I never learned to code. Every line in Penglai was written by me speaking to AI coding tools, one sentence at a time.
Penglai is the very thing it set out to prove: in the AI era, ordinary people can build their own tools.

Why "Penglai"? In Chinese mythology, Penglai is a legendary island of immortals on the sea β€” beautiful, yet forever out of reach. This project reclaims that name to say: those seemingly distant AI capabilities can actually be within reach β€” installed on your own machine, driven by the tools you already know.

This is a Penglai that is within reach.

Eight Immortals Cross the Sea

Multiple channels, capabilities, and models working in harmony β€” each module crosses the sea its own way, serving the same you

Desktop Feishu WeChat Terminal DingTalk Runtime Hub GenericAgent Voice Stack Memory Safety CHANNELS CORE CAPABILITIES
One command to add a new channel
# Enable Feishu channel (scan QR to auto-create app)
penglai enable feishu
 
# Start Runtime Hub service
penglai start
 
# Enable proactive companion mode
penglai companion enable

Numbers That Speak

Not features on a roadmap β€” capabilities running daily on real servers

0
Channels
Desktop, Feishu, WeChat, Terminal, DingTalk, QQ, WeCom, Telegram
0
Memory Layers
Index, Fact, Skill, Session β€” fully Markdown-auditable
0
Data Leaks
Local CPU voice inference, data stays on your machine by default
0
Minutes to Start
Scan QR to connect, one command, up and running in ten minutes
πŸŽ™οΈ

Local Voice Dual-Stack

SenseVoice speech recognition + MOSS-TTS speech synthesis, all running on your local CPU. No GPU, no cloud services β€” just say a word to interact, without touching anything beyond microphone permissions.

Learn more
πŸ›‘οΈ

Deterministic Safety Guardrails

A rule-based safety layer that does not rely on AI judgment. Tool-call whitelists, tiered permissions, operation audit logs β€” every step is deterministic logic, reviewable and accountable.

Learn more
🧐

Anti-Hallucination Double-Safety

A four-layer memory architecture ensures the agent "knows" what it knows and what it doesn't. The fact layer stores verified information, the index layer provides contextual retrieval, effectively reducing AI-fabricated content.

Learn more
πŸŒ™

Proactive Without Being Annoying

Not a passive bot that replies to every message. Penglai Agent can proactively inspect systems, send scheduled reports, and trigger anomaly alerts, while strictly respecting do-not-disturb hours and frequency limits β€” working for you without bothering you.

Learn more

One Butler, Many Doors

Your AI Agent should live wherever you already open every day

Desktop Client
Verified
πŸŽ™οΈ Voice support
Download Desktop
Feishu
Verified
πŸŽ™οΈ Voice support
penglai enable feishu
WeChat
Verified
πŸŽ™οΈ Voice support
penglai enable wechat
Terminal
Verified
πŸŽ™οΈ Voice support
penglai
DingTalk
⚠️ Pending
QQ
⚠️ Pending
WeCom
⚠️ Pending
Telegram
⚠️ Pending

Get Started

Three methods β€” pick whichever suits you

1

Download the 0.3.6 installer for your platform (macOS Apple Silicon / macOS Intel x64 / Windows x64). Desktop packages include the Penglai runtime, standalone Python, and core dependencies.

https://github.com/kevinchennewbee/PenglaiAgent/releases/latest
2

Open the installer β€” drag to Applications (Mac) or follow the setup wizard (Windows). First launch no longer requires manual Python, source checkout, or live pip downloads.

3

Launch Penglai Desktop and follow the wizard to configure your first Agent. In-app update metadata now defaults to mainland-friendly gh-proxy asset URLs.

Open Penglai β†’ Click "Get Started"
1

Install Penglai CLI with one command (auto-provisions Python & deps)

curl -fsSL https://raw.githubusercontent.com/kevinchennewbee/PenglaiAgent/main/install.sh | sh
2

Start Runtime Hub service (Feishu + WeChat + scheduler)

penglai start
3

Start chatting (direct TUI, equivalent to upstream GA tuiapp_v2)

penglai
1

Install Penglai core via pip (PyPI package name: penglai)

pip install penglai
2

Run the setup wizard (auto-configures LLM, identity, channels, abilities)

penglai setup
3

Configure individual modules as needed (e.g., Feishu only)

penglai setup --only feishu

Recent 0.3.x Release Trail

The website keeps the recent 0.3.x release history visible so users can see how Penglai evolved from Runtime Hub to a desktop-ready distribution.

🌌

v0.3.6 Β· 2026-07-12

Verifiable Worldline release: conversation rewind is completed as a persistent branching timeline, with working-memory restoration and native-log reconciliation when a session is resumed. This release also fixes long-prompt temporary-file races, broken browser pipes, Claude refusal finalization, and Companion failure outcome tracking, while hardening three-platform release gates, version consistency, and Rust download fallback.

πŸ”

v0.3.5 Β· 2026-07-03

Trusted release gates and Companion Loop foundation: the three-platform desktop release keeps real installer validation while adding signed release-tag verification, GitHub Release artifact aggregation, latest.json updater output, and the policy / event / memory interface skeleton for Companion Loop. Companion Loop is a verifiable foundation here, not a claim of a completed production adaptive loop.

πŸ›‘οΈ

v0.3.4 Β· 2026-06-30

Trusted-maintenance release: keeps the three-platform desktop distribution from 0.3.3 while fixing the desktop one-click update path, Runtime Hub cancellation, Conductor authentication, inline eval exposure, memory load-time scanning, untrusted external tool-result marking, WeChat / WeCom media path sanitization, LLM failover classification, and fail-closed updater metadata checks for missing bundles or signatures.

🏁

v0.3.3 Β· 2026-06-29

Intel Mac desktop support release: adds a dedicated macOS Intel x64 DMG and darwin-x86_64 updater metadata while keeping the 0.3.2 bundled runtime model. First-run setup now includes Hermes / OpenClaw migration preview, avoids returning plaintext secrets by default, and requires a second confirmation token for state-changing operations.

πŸ“¦

v0.3.2 Β· 2026-06-28

Local desktop runtime release: macOS Apple Silicon DMG and Windows x64 installer bundle the Penglai runtime, standalone Python, and core dependencies. First launch no longer asks users to install Python, clone the source tree, or download Python packages live. latest.json defaults to gh-proxy.com release asset URLs.

πŸ”

v0.3.1 Β· 2026-06-27

Migration, auto-update, and dynamic versioning milestone: adds penglai backup / restore / uninstall-legacy, setup --only partial reconfiguration, and a six-gate tauri-plugin-updater path. Runtime Hub stability, companion modes, and context closure were tightened.

🧭

v0.3.0 Β· 2026-06-25

Runtime Hub GA: native Mac / Windows desktop clients, Feishu QR auto-create, MOSS-TTS-Nano local TTS, penglai update with automatic backup and rollback, and Docker removed from the public support matrix.

πŸ§ͺ

v0.3.0-preview.1 Β· 2026-06-25

Real 0.3.0 preview artifact validation: GitHub prerelease produced both macOS DMG and Windows NSIS artifacts, Windows silent install / uninstall verification passed, and Tencent Cloud install-check, runtime-audit, privacy-audit, and service-state verification all passed.

πŸ›‘οΈ

Release Boundary

Known limit: macOS and Windows installers are not signed with Apple Developer ID / Microsoft Authenticode yet, so first launch may still show Gatekeeper / SmartScreen warnings. Choose the Apple Silicon or Intel x64 DMG that matches your Mac.

Honesty over aesthetics

Every feature above runs on real servers daily β€” not a roadmap.
Every capability in Penglai is something you can verify right now.

GenericAgent
SenseVoice
MOSS-TTS-Nano
Tauri