Skip to main content
Pre-release

This release is a pre-release and may not be stable for production use.

psi-agent

English

用 Python socket 拼装 AI agent 的微框架。

你只需要写 Python 函数和 Markdown——剩下的 socket 通信、tool calling、SSE 流式、定时任务框架全替你做了。

为什么选它

  • 简单组合:ai、session、channel 三个独立进程,socket 对插即用。没有中心配置,不依赖数据库
  • Workspace 即 Agent:在 workspace/ 下丢几个 Python 函数就是 tools,写个 system prompt 就是 agent 人格,加个 cron 就是定时任务
  • 流式交互:REPL 实时显示 AI 思考过程(dim 样式)和最终回复,所见即所得

架构

用户 ←→ Channel (REPL/CLI) ──Unix socket── Session ──Unix socket── AI (OpenAI/Anthropic)

快速开始

需要 Python >= 3.14

三个终端,三步跑起来:

# 安装
uv sync

# 终端 1:启动 AI 后端(--api-key 可选,不传则读 PSI_AI_API_KEY 环境变量)
uv run psi-agent ai \
  --provider openai \
  --session-socket ./ai.sock \
  --model gpt-4o-mini \
  --api-key $OPENAI_API_KEY \
  --base-url https://api.openai.com/v1

# 终端 2:启动 Session
uv run psi-agent session \
  --workspace ./examples/a-simple-bash-only-workspace \
  --channel-socket ./channel.sock \
  --ai-socket ./ai.sock

# 终端 3:REPL 交互
uv run psi-agent channel repl --session-socket ./channel.sock

REPL 操作:Enter 换行,Alt+Enter 发送,Ctrl+D 退出。

也可以一句命令搞定:

uv run psi-agent channel cli \
  --session-socket ./channel.sock \
  --message "列出当前目录的文件"

CLI 一览

psi-agent
├── ai                        # 统一 AI 后端(支持 50+ provider)
├── session                    # Session + workspace 管理
└── channel
    ├── repl                   # 交互式 REPL
    └── cli                    # 单次消息

定义你自己的 Agent

一个 workspace 就是一个 agent:

my-workspace/
├── tools/                    # 每个 .py 文件定义一个 tool
│   └── bash.py               # async def bash(command: str) -> str
├── skills/                   # */SKILL.md 技能文档(system_prompt_builder 自行遍历)
├── schedules/                # 定时任务
│   └── daily-report/
│       └── TASK.md           # YAML 头 (name, cron) + Markdown body
└── systems/
    └── system.py             # async def system_prompt_builder() -> str

Tool 就是一个 async 函数,函数名 = 文件名,参数类型自动映射为 JSON Schema:

# tools/bash.py
import anyio

async def bash(command: str) -> str:
    """Execute a bash command.
    Args:
        command: The command to run.
    """
    result = await anyio.run_process(["/bin/bash", "-c", command])
    return result.stdout.decode().strip()

定时任务的 body 会在 cron 触发时作为消息发送给 AI:

---
name: daily-report
cron: "0 12 * * *"
---
请生成一份项目进展日报。

更多细节见 examples/a-simple-bash-only-workspace/。

开发

uv run ruff check .          # lint
uv run ruff format --check . # 格式
uv run ty check              # 类型
uv run pytest -v             # 测试

作者

Hao Zhang hzhangxyz@outlook.com

许可

MIT License. 详见 LICENSE。

Metadata

Release files for psi-agent 0.0.1a20260623

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for psi-agent 0.0.1a20260623
File Size Uploaded Provenance Actions
psi_agent-0.0.1a20260623.tar.gz 102.4 kB With provenance Details for psi_agent-0.0.1a20260623.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for psi-agent 0.0.1a20260623
File Interpreter ABI Platform Actions
psi_agent-0.0.1a20260623-py3-none-any.whl Python 3 none any Details for psi_agent-0.0.1a20260623-py3-none-any.whl

Total release size: 132.6 kB

Release files / psi_agent-0.0.1a20260623.tar.gz

Download URL psi_agent-0.0.1a20260623.tar.gz
Size 102.4 kB
Tags Source
SHA-256 checksum
How to use checksums
46477129386fcc551d6637d4f2614b35b794ca3c60c58d6ab3d0b593322384ae
BLAKE2b-256 checksum
How to use checksums
072372b7daca7c1d3cab2b7b4e5d5f55e00e62e5474b0fb37bf67dfbd15237ec
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/6.1.0 CPython/3.13.12

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Jun 23, 2026.

Transparency log

Release files / psi_agent-0.0.1a20260623-py3-none-any.whl

Download URL psi_agent-0.0.1a20260623-py3-none-any.whl
Size 30.2 kB
Tags Python 3
SHA-256 checksum
How to use checksums
e3cac34cdef7b77460f23cae65a269556e9c2e7880b825d55053d409f9af380a
BLAKE2b-256 checksum
How to use checksums
5b57763ffef1f397810efb186511dbc686f033ec68bd9a21bc722e451d05fb95
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/6.1.0 CPython/3.13.12

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Jun 23, 2026.

Transparency log

Security for psi-agent 0.0.1a20260623

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

All 2 files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for psi_agent-0.0.1a20260623.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Jun 23, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release
Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page