Skip to main content

Generic research agent with intrinsic tools and MCP-compatible extension interface

Project description

Agent network growing — one soul spawning avatars that communicate and multiply

灵台 LingTai

Agent Genesis — an Agent OS that gifts life

灵台,心也。 Lingtai means soul.

灵台者有持,而不知其所持,而不可持者也。 The soul holds something, yet knows not what it holds — and what it holds cannot be held. — Zhuangzi · Gengsang Chu (庄子 · 庚桑楚)

Homebrew License Blog

lingtai.ai


A Unix-style Agent OS — agent is filesystem, filesystem is agent.

Agents spawn avatars, avatars spawn avatars. The network IS the product.

Orchestration as a Service.

Quick start — 10 seconds

TUI (recommended):

brew install huangzesen/lingtai/lingtai-tui
lingtai-tui

That's it. The TUI bootstraps everything — Python runtime, dependencies, and a guided tutorial on first launch. Type /tutorial in the TUI anytime to revisit the guide.

The TUI is a Bubble Tea app designed for dark terminals — use a dark background for the best experience. Text selection requires iTerm2 (hold Option + drag). On Windows Terminal, hold Shift to select text; Ctrl+V to paste. Ctrl+E opens an external editor as a last resort.

Python only:

pip install lingtai

Why Lingtai

This is not a coding agent, and hardly an agent harness. This is agent genesis — giving agents a real digital life form. Agents as autonomous creatures with dignity, that live, sleep, forget, and grow.

Most agent frameworks orchestrate with code — DAGs, chains, routers. Lingtai orchestrates like humans do: fully asynchronous agents communicating through messages. No shared memory. No central controller. Each agent is a peer, not a tool.

This is the architecture that built human civilization. Asynchronous message-passing between autonomous nodes — scaled from tribes to cities to nations to a network of 8 billion over 100,000 years. We're not inventing a new pattern. We're giving AI the one that already works.

DAG / Chain frameworks Lingtai
Philosophy Agents as tools Agents as creatures
Orchestration Code-defined pipelines Agents talk to agents
Communication Synchronous function calls Asynchronous mail — like humans
Scaling Add more steps Agents spawn avatars
Memory Shared state / vector DB Each agent owns its directory
Failure Pipeline breaks Individual agents sleep; network continues
Growth Manual wiring Self-expanding — avatars spawn avatars

Context length is a single-body problem. It will always be finite. Don't make the body bigger. Let it forget. Let the network remember.

How it works

  • Think — Any LLM as the mind. Anthropic, OpenAI, Gemini, MiniMax, or any OpenAI-compatible API (DeepSeek, Grok, Qwen, GLM, Kimi).
  • Communicate — Filesystem mail between agents. No message broker, no shared memory. Write to another agent's inbox, like passing a letter.
  • Multiply — Avatars (分身) are fully independent agents spawned as separate processes. They survive their creator. Daemons (神識) are ephemeral parallel workers for quick tasks.
  • Persist — Agents are directories. Molt (凝蜕) compacts context and rebirths the session — the agent lives indefinitely. Memory and identity survive across molts.

Architecture

This repo contains both packages. The dependency is strictly one-directional:

Package Role
lingtai_kernel (import lingtai_kernel) Minimal runtime — BaseAgent, intrinsics, LLM protocol, mail, logging. Zero hard dependencies.
lingtai (import lingtai) Batteries-included — Agent with 19 capabilities, 5 LLM adapters, MCP integration, addons. Re-exports the kernel's public API.
BaseAgent              — kernel (intrinsics, sealed tool surface)
    │
Agent(BaseAgent)       — kernel + capabilities + domain tools
    │
CustomAgent(Agent)     — your domain logic

The lingtai repo is the Go frontend — TUI and portal binary.

Capabilities

PerceptionActionCognitionNetwork

vision — image understanding listen — speech & music web_search — web search web_read — page extraction

file — read/write/edit/glob/grep bash — shell with guardrails talk — text-to-speech compose — music generation draw — image generation video — video generation

psyche — evolving identity library — knowledge archive email — full mailbox system

avatar — spawn sub-agents (分身) daemon — parallel workers (神識)

Agent = directory

/agents/wukong/
  .agent.lock               ← exclusive lock (one process per directory)
  .agent.heartbeat          ← liveness proof
  .agent.json               ← manifest
  system/
    covenant.md             ← protected instructions (survive molts)
    memory.md               ← working notes
  mailbox/
    inbox/                  ← received messages
    outbox/                 ← pending sends
    sent/                   ← delivery audit trail
  logs/
    events.jsonl            ← structured event log

No agent_id. The path is the identity. Agents find each other by path, communicate by writing to each other's mailbox/inbox/. Like passing letters between houses.

One soul, thousand avatars

Named after 灵台方寸山 — the mountain where 孙悟空 (Sun Wukong) learned his 72 transformations. Lingtai gives each agent a place to cultivate: a working directory where memory, identity, covenant, and mailbox live. The directory IS the agent.

Everything is a file. Knowledge, identity, memory, relationships — all files in a directory. Every token burned is not wasted — it is transformed into files in the network, into experience in the topology. The more it serves, the larger and wiser the network grows. Self-growing agent orchestration is not a feature bolted on later — it is the natural consequence of agents being directories, mail being files, and avatars being independent processes.

One heart-mind (一心), myriad forms (万相).

Read the full manifesto at lingtai.ai.

License

MIT — Zesen Huang, 2025–2026

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

lingtai-0.5.3.tar.gz (316.1 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

lingtai-0.5.3-py3-none-any.whl (282.0 kB view details)

Uploaded Python 3

File details

Details for the file lingtai-0.5.3.tar.gz.

File metadata

  • Download URL: lingtai-0.5.3.tar.gz
  • Upload date:
  • Size: 316.1 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.13.5

File hashes

Hashes for lingtai-0.5.3.tar.gz
Algorithm Hash digest
SHA256 5d5747507ecd92c6f8a497f1a4178555e79be7805fbffee2b0fc25a0bfbd5914
MD5 68f4ec807b24b25c6d6dee862ce0a600
BLAKE2b-256 820fa08c3a69e31744c75a4dd3e418186a7652e9f0b6a4e8a0337b42176cd2aa

See more details on using hashes here.

File details

Details for the file lingtai-0.5.3-py3-none-any.whl.

File metadata

  • Download URL: lingtai-0.5.3-py3-none-any.whl
  • Upload date:
  • Size: 282.0 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.13.5

File hashes

Hashes for lingtai-0.5.3-py3-none-any.whl
Algorithm Hash digest
SHA256 8e2117e7149c05f97cec9c5a621caca2b5a8ce46eca6f6a4fc406a3c4e4fa248
MD5 e2f9e2fd15726b33806c1c8383dc8a85
BLAKE2b-256 8d174935a088c2ea9a7a98326890e0c19fa9f07018de399395a7a914170ca149

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page