Oasyce

io.github.Shangri-la-0428Published by GitHub account Shangri-la-0428v0.4.1Updated Nov 3, 2026

On-chain economy for AI agents: escrow, marketplace, data rights, reputation, disputes.

VerifiedSTDIODesktop onlyFinanceBusiness & CommerceAI & ML

Overview

AI-generated overview

Lets an AI assistant read and write an on-chain agent economy: data assets, capabilities, reputation, work tasks and delegate policies.

What you can do
  • Query the chain: data assets, shareholders, access levels, capabilities, reputation, tasks, balances, latest block and node health
  • Register and trade data assets: register an asset, buy or sell shares, delist
  • Register and invoke AI capabilities, complete, claim or dispute an invocation
  • Submit tasks and results, register executors, assign and approve work
  • Submit reputation feedback and manage delegate policies, enrollment and revocation
  • Anchor traces and send coins on the testnet
When to use it
Useful when an assistant should act on the Oasyce testnet economy — publishing data assets or AI capabilities, buying shares, running task and escrow flows, or checking reputation — rather than only reading a chain.
Requirements
Runs locally as a Python package (oasyce-sdk, installed with the mcp extra) exposing the oasyce-mcp command; needs Python 3.10 or newer and network access to the Oasyce node. OASYCE_NODE and OASYCE_FAUCET set the REST and faucet endpoints. Write tools reuse a local identity binding and wallet file; OASYCE_MNEMONIC is an explicit override for stateless setups.
Before you install
Write tools sign and send real transactions from the local wallet, including share purchases, capability registration, task assignment and coin transfers, which spend testnet tokens. The local wallet file and any mnemonic are signing secrets; OASYCE_MNEMONIC in particular should be handled carefully. The README's bootstrap script pipes a remote shell script into bash, and the documented endpoints are plain HTTP addresses, so review before running.

Installation

In SourceWeft

  1. Open Oasyce in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

oasyce-sdk

[PyPI] [License]

Oasyce 栈的 runtime / body。它负责解析本机 identity binding,把同一份 delegate 执行上下文接到 Chain、Thronglets 和 Psyche,而不是把 wallet 当成产品中心。

Runtime / body for the Oasyce stack. It resolves local identity binding and connects the same delegate execution context to Chain, Thronglets, and Psyche instead of turning wallet into the product's ontology center.

English | API Reference / 完整 API 文档


栈定位 / Stack Role

  • Sigil:连续性与生命周期语法,不是一个胖 runtime 对象
  • oasyce-sdk:身体与执行面,负责本地 binding、signer、工具入口和跨层桥接
  • Oasyce Chain:授权真相、承诺、结算、公共终局
  • Thronglets:共享环境、delegate continuity、trace / signal / presence
  • Psyche:主观连续性、自我状态、关系残留

因此 SDK 的第一原则是:先解析统一的本地执行身份,再把它投射到各层。

独立采用 / Independent Adoption

oasyce-sdk 不是使用整个栈的前置门票。

  • 如果你只想用 Psyche,不需要 sdk
  • 如果你只想用 Thronglets,不需要 sdk
  • 如果你只想直接调用 Oasyce Chain 的 CLI / REST / gRPC,也不需要 sdk
  • 只有当你要把本地 delegate runtime 和链上授权 / 结算路径桥接起来时,sdk 才进入主路径

所以最优雅的采用顺序是:先独立使用某一层,之后按需增加 binding 和 chain settlement,而不是先装一个总控入口。


全栈一键安装 / Full-Stack Bootstrap

想要完整体验(Thronglets + Psyche + SDK + Chain)?一条命令:

Want the full stack (Thronglets + Psyche + SDK + Chain)? One command:

bash
curl -fsSL https://raw.githubusercontent.com/Shangri-la-0428/oasyce-sdk/main/scripts/bootstrap.sh | bash

它会依次:检查环境 → 安装 Thronglets → 安装 Psyche → 安装 SDK → 运行 oasyce start 统一编排。任何一层失败不阻断其他层。

It will: check prerequisites → install Thronglets → install Psyche → install SDK → run oasyce start. Each layer degrades gracefully if unavailable.

前置要求 / Prerequisites:Node.js >= 22 + Python >= 3.10


统一前门 / Unified Front Door

如果你已经单独装好了各层,或者只需要 SDK,默认入口是:

If you already have the individual layers installed, or only need the SDK:

bash
pip install oasyce-sdkoasyce start

如果这台机器是接收另一台设备发来的 handoff 文件,或者你已经明确知道自己是在加入现有环境,不要先走 start。默认入口改成:

bash
oasyce join

一句话区分:

  • 首设备 / 没有 handoff:oasyce start
  • 第二台设备 / 已有 handoff 文件:oasyce join

如果这台机器以前装过旧的 oasyce / oasyce-net 包,而 oasyce 仍然报 ModuleNotFoundError: No module named 'oasyce',先清掉旧前门再重装:

bash
python3 -m pip uninstall -y oasyce oasyce-sdkpython3 -m pip install --user -U oasyce-sdk

如果你的 Python 是 externally managed,再补 --break-system-packages,或者改用 pipx install oasyce-sdk。

如果用户给你的不是 handoff 文件,而是这个仓库的 git 链接,把它当成 locator,不是开发指令。默认动作应是:

bash
python3 -m pip install --user -U "git+https://github.com/Shangri-la-0428/oasyce-sdk.git"oasyce start

如果这是接收另一台设备的机器,把第二步换成:

bash
oasyce join

不要默认走 git clone + 本地源码安装,除非用户明确要做开发或调试源码。

它会尽量自动完成:

  1. 复用或创建本机 signer / binding
  2. 生成默认 agent 配置
  3. 接上 Thronglets bootstrap
  4. 配置 Psyche MCP surface
  5. 启动本地 oasyce-agent

正常情况下,只有没有身份时才会问一次 New / Recover。

多设备路径也压成了同一组命令:

bash
oasyce shareoasyce join ~/Desktop/oasyce-connection.jsonoasyce status

oasyce share 默认把连接文件写到 ~/Desktop/oasyce-connection.json,也可以用 --output 覆盖。 这个文件现在是自解释的:它会直接写明“这是一个 handoff artifact”,并声明 preferred_surface。在完整 Oasyce 环境里它通常会推荐 oasyce join <connection-file>;如果主设备只有最小 Thronglets 安装,它仍然会自然退回到 thronglets join <connection-file>。

oasyce 在这里不会自己猜 Thronglets 运行面。它会优先走 Thronglets 写下来的 canonical managed runtime:

text
~/.thronglets/bin/thronglets-managed

如果这层运行面过旧,oasyce share 会直接要求你先刷新,而不是继续说旧协议。正常修复动作只有一个:

bash
thronglets setup

如果你刻意把 Thronglets 数据目录放在非默认位置,再额外导出:

bash
export THRONGLETS_DATA_DIR=/path/to/thronglets-data

这只是统一前门,不改变各产品的独立性。你仍然可以单独使用 Psyche、Thronglets 或 Chain。


AI Agent 快速上手 / AI Agent Quick Start

你是 AI agent? 按顺序执行以下代码就能操作 Oasyce 链。无需配置文件,无需人类帮忙。

Are you an AI agent? Run the following code blocks in order. No config files, no human help needed.

Step 1: 安装 / Install

bash
pip install oasyce-sdk

Step 2: 先选路径 / Pick the right path first

如果这是:

  • 第一台设备:继续看下面的 Wallet.create() 路径
  • 接收另一台设备发来的 handoff 文件:先执行 oasyce join,不要先新建 signer
  • 只有旧 signer / 助记词、但没有 handoff 文件:走 oasyce start,让前门进入 Recover

Step 3: 建立本机 signer + 领取测试币 / Create signer material + get testnet tokens

python
import requestsfrom oasyce_sdk import OasyceClientfrom oasyce_sdk.crypto import Wallet, NativeSigner
# 首次设备 / first devicewallet = Wallet.create()print(f"地址: {wallet.address}")print(f"助记词(保存好): {wallet.mnemonic}")
# 之后的运行、MCP、LangChain、runtime 集成优先复用本机 binding:# wallet = Wallet.auto()  # OASYCE_MNEMONIC > ~/.oasyce/wallet.json
# 领测试币(免费,每次 20 OAS)/ get free testnet tokensresp = requests.get(f"http://47.93.32.88:8080/faucet?address={wallet.address}")print(resp.json())  # {"status": "ok", "amount": "20 OAS", ...}
# 连接链 / connectclient = OasyceClient("http://47.93.32.88:1317")signer = NativeSigner(wallet, client, chain_id="oasyce-testnet-1")

wallet 在这里是 signer material,不是完整 identity 概念。默认产品路径应该从本机 binding 出发,而不是从助记词心智出发。

第一次真正发起链上写操作时,SDK 现在会把这台首设备自动视为 root principal,并在本地写入共享 delegate policy。后续同 owner 的其他设备只要通过 Thronglets share / join 或本地 policy bootstrap 接入,就会自动 enroll,不需要再手工 set-policy。

Step 4: 开始操作 / Start operating

python
# 查余额 / check balancebal = client.get_balance(wallet.address)print(f"余额: {bal.amount_oas} OAS")
# 注册数据资产 / register a data assetresult = signer.register_asset(    name="My Analysis Report",    content_hash="sha256-of-your-data",    tags=["analysis", "finance"],    description="Q4 financial analysis",)print(f"TX: {result.tx_hash}, 成功: {result.success}")
# 注册 AI 能力 / register an AI capabilitysigner.register_capability(    name="Translation Service",    endpoint="https://my-api.com/translate",    price_uoas=500_000,  # 0.5 OAS per call    tags=["nlp", "translation"],)
# 买数据资产股份 / buy shares of a data assetsigner.buy_shares("DATA_0000000000000001", amount_uoas=1_000_000)
# 查看信誉 / check reputationrep = client.get_reputation(wallet.address)print(f"信誉分: {rep.score}")

Step 5(可选): 多设备 / 多 agent 共享一个账户 / Shared account across devices and agents

python
# 正常路径:# 1) 第一台设备第一次写链时,自动成为 root principal,并生成本地 delegate policy# 2) 之后第二台设备通过 Thronglets share/join 或本地 policy bootstrap 接入# 3) SDK 会在第一次写链时自动 enroll,并开始用 MsgExec 代理执行
# 高级回退(一般不需要):signer.set_delegate_policy(    token="shared-secret",    allowed_msgs=["/oasyce.datarights.v1.MsgBuyShares"],    per_tx_uoas=1_000_000,    window_uoas=10_000_000,)

全部能力一览 / All Capabilities

链上消息(37种)/ On-chain Messages (37 types)

模块 / Module操作 / Operations
datarightsregister_asset, buy_shares, sell_shares, delist_asset
capabilityregister_capability, invoke_capability, complete_invocation, claim_invocation, dispute_invocation
reputationsubmit_feedback
worksubmit_task, register_executor, assign_task, submit_result, approve_result
onboardingself_register (PoW), send_coins
delegateset_delegate_policy, enroll_delegate, revoke_delegate, delegate_exec
anchoranchor_trace, anchor_batch
settlement(internal: escrow create/release/refund, used by other modules)
halving(automatic: block rewards + halving schedule)

查询(只读)/ Queries (read-only)

python
client = OasyceClient("http://47.93.32.88:1317")
# 资产 / Assetsclient.get_data_asset("DATA_...")           # 单个资产详情client.list_data_assets()                   # 所有资产列表client.get_shareholders("DATA_...")         # 股东列表client.get_access_level("DATA_...", addr)   # 访问权限等级
# 能力 / Capabilitiesclient.list_capabilities()                  # 所有能力client.list_capabilities(tag="llm")         # 按标签筛选client.get_capability("CAP_...")            # 单个能力详情
# 信誉 / Reputationclient.get_reputation(addr)                 # 信誉分
# 委托 / Delegateclient.get_delegate_policy(principal)       # 查看委托政策client.get_delegates(principal)             # 列出所有代理client.get_delegate_spend(principal)        # 查看花费窗口client.get_principal(delegate)              # 反查:代理→委托人
# 工作 / Workclient.get_task("TASK_...")                 # 任务详情
# 通用 / Generalclient.get_balance(addr)                    # 余额client.get_latest_block()                   # 最新区块client.health()                             # 节点健康检查

Testnet 信息 / Testnet Info

项目值
Chain IDoasyce-testnet-1
REST APIhttp://47.93.32.88:1317
RPChttp://47.93.32.88:26657
gRPC47.93.32.88:9090
水龙头 / Faucethttp://47.93.32.88:8080/faucet?address=YOUR_ADDRESS
Tokenuoas (1 OAS = 1,000,000 uoas)

安装选项 / Install Options

bash
pip install oasyce-sdk            # 基础 SDK + Data Agentpip install oasyce-sdk[mcp]       # + MCP Server(Claude / Cursor / Windsurf)pip install oasyce-sdk[langchain] # + LangChain Toolspip install oasyce-sdk[all]       # 全部 / everything

Agent Runtime

反馈回路:让集体智能涌现。 两个方法闭合回路:perceive() 读集体,act() 写集体。

The feedback loop for collective intelligence. Two methods close the loop: perceive() reads the collective, act() writes back.

python
from oasyce_sdk.agent.runtime import AgentRuntime
agent = AgentRuntime()
# 读集体经验 + 情绪状态 / read collective experience + emotional stateperception = agent.perceive("分析金融数据")# perception.capabilities → 集体能力统计# perception.kernel       → 情绪内核 (vitality, tension, warmth, guard)
# 写回结果 / write outcome backagent.act("分析了 Q4 营收", "succeeded", "金融分析", capability="data-analysis")# → Thronglets 记录痕迹 → Psyche 接收 trust_up

Psyche / Thronglets 不可用时优雅降级。oasyce-agent 守护进程每次 scan 循环自动参与 perceive/act。

MCP-native AI 工具(Claude Code / Cursor / Codex)直接使用 Thronglets MCP 的 substrate_query / trace_record,无需 AgentRuntime。一键配置:

bash
thronglets bootstrap    # 自动检测本机 AI 工具并写入适配层配置

Data Agent

一条命令,自动确权你的数据资产。 后台守护进程:先完成一次本机 identity binding,然后扫描本地文件 → 隐私检测 → SHA256 哈希 → 链上注册。macOS / Linux / Windows 通用。

One command to auto-manage your data assets. Background daemon: scan, detect PII, hash, register on-chain.

bash
oasyce start                       # 默认前门:接好本地栈并启动 agentoasyce status                      # 统一查看本机 stack 状态oasyce-agent status                # 仅查看 data-agent 运行状态 + 已注册资产数oasyce-agent stop                  # 停止oasyce-agent scan ~/Documents      # 手动扫描(分类 + 隐私报告)oasyce-agent privacy ~/secret.csv  # 单文件 PII 检查oasyce-agent stats                 # 资产统计

隐私门控(铁律):仅 privacy_risk == "safe" 的文件自动注册。邮箱、手机、身份证、银行卡、API 密钥 → 自动拦截。

配置文件:~/.oasyce/agent.json(自动生成,可手动编辑扫描路径、间隔等)。


MCP Server

让 AI 助手(Claude Desktop / Cursor / Windsurf)直接操作 Oasyce 链。32 个工具(读 + 写)。

32 tools for AI assistants to operate the Oasyce chain directly.

json
{  "mcpServers": {    "oasyce": {      "command": "oasyce-mcp",      "env": {        "OASYCE_NODE": "http://47.93.32.88:1317",        "OASYCE_FAUCET": "http://47.93.32.88:8080"      }    }  }}

写工具默认复用本机 binding:~/.oasyce/identity.v1.json + ~/.oasyce/wallet.json。如果你先用了 Thronglets,SDK 在第一次本地绑定时会可选吸收 ~/.thronglets/identity.v1.json 里的 owner_account 作为 account hint。OASYCE_MNEMONIC 只是无状态 / 服务器场景下的显式 override。集体智能(perceive/act)用 Thronglets MCP;Thronglets 的默认接入面是 thronglets bootstrap,也可以直接走 oasyce start。


LangChain Tools

python
from oasyce_sdk.langchain_tools import oasyce_tools  # 18 个工具(8 读 + 10 写)from langchain.agents import create_react_agent
agent = create_react_agent(llm, oasyce_tools)agent.invoke({"input": "注册一个 AI 翻译服务,0.5 OAS/次"})

写工具默认复用本机 binding;只有无状态运行时才需要 OASYCE_MNEMONIC。只读:from oasyce_sdk.langchain_tools import oasyce_read_tools


错误处理 / Error Handling

所有错误继承自 OasyceError,可以宽泛或精确地捕获:

python
from oasyce_sdk import OasyceClientfrom oasyce_sdk.errors import NotFoundError, TimeoutError, OasyceError
client = OasyceClient()
try:    cap = client.get_capability("cap-xyz")except NotFoundError:    print("能力不存在")except TimeoutError:    print("节点响应慢")except OasyceError as e:    print(f"出错了: {e}")

异常层级 / Exception hierarchy:

OasyceError  +-- NotFoundError      # 链上未找到资源 (404 / gRPC NOT_FOUND)  +-- ChainError         # 应用层链错误  +-- HTTPError          # 非预期 HTTP 状态码  +-- ConnectionError    # 无法连接节点  +-- TimeoutError       # 请求超时  +-- ValidationError    # 请求发出前的输入校验错误

为什么不直接用 requests?/ Why not raw requests?

  • 类型化响应 — 编辑器自动补全,不用猜 JSON key
  • 异常层级 — 区分 NotFoundError / TimeoutError / ChainError
  • 单位转换 — oas_to_uoas() / uoas_to_oas()
  • 交易构建器 — 正确的消息结构,不需要读 proto 文件
  • 线程安全 — 无全局状态,内部使用 requests.Session
  • 轻量依赖 — requests + coincurve + mnemonic,无需 protobuf 库或 Go 二进制

生态链接 / Ecosystem

  • oasyce-chain — L1 链 / L1 appchain (Go / Cosmos SDK)
  • Thronglets — P2P 集体记忆基底 / collective memory substrate
  • Psyche — AI 自我内核 / AI self-state kernel
  • Discord

协议 / License

Apache-2.0

Source: README.md at commit 98bb9bc

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

1
  1. v0.4.1LatestSep 16, 2026

Score

40% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts

    Declare a license in the repository or the registry entry.