Vibedonate

io.github.pooriaarab由 GitHub 账号 pooriaarab 发布v0.4.2更新于 Oct 24, 2026

BitTorrent-style mesh for AI inference across agentic CLIs. Local-first.

发布方自己的服务:AI Benchmark Index

概览

AI 生成的概览

让助手查看本机当前是否正在向对等网络捐赠闲置本地算力,并对容量请求做预检。

可以做什么
以本地 stdio 方式运行 MCP 服务器,提供两个工具:status 报告捐赠配置、计量总量以及节点当前是否正在共享;request_capacity 接收对等方标识与 token 数量,返回 allow 或 deny 及原因。容量检查只是预检:它会核对 donate:compute 同意授权、对等方授权、本机模型可用性、空闲时间窗、本机活动与每日 token 上限,但不消耗容量。捐赠本身通过 vibedonate 命令行启用,该命令行还提供 status、stop 和库接口。
适用场景
如果你已经在使用 vibedonate 命令行,并希望编码助手了解本节点是否在共享、某个对等方的请求是否会被放行,可以添加。单独使用价值有限,因为该 MCP 服务器只做报告与预检,不会启动或配置捐赠。
运行要求
需要本地 Node.js 运行环境,因为它是以 npx 运行的 npm 包,也可全局安装。未声明任何账号、API 密钥或请求头。算力检查需要本机模型运行器;README 提到通过环境变量 VIBEDONATE_OLLAMA_MODEL 配置的 Ollama 运行器。状态保存在 VIBEDONATE_DIR 下,默认为 ~/.vibedonate。
安装前请注意
该服务器只是预检闸门,本身不处理远程请求;README 说明 P2P 传输尚未落地,因此此版本不承载远程对等方流量。捐赠为选择性加入,并按每日 token 上限与对等方池计量,README 称本机解析禁止外发,捐赠算力留在本机。启用捐赠前请检查同意授权与池设置,并注意 README 明确表示本版本不包含把陌生人的推理经由你的订阅或 API 密钥路由。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Vibedonate,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

vibedonate

[npm version] [Downloads/month] [License: MIT]

Package · Setup · Runtime surfaces · Compare

vibedonate is an open-source, self-hosted compute-sharing experiment for developers who use coding agents. Your hardware, opt-in, metered. It provides peer transport, donation controls, and local ledgers in the Vibe Suite. The current CLI mesh runs an echo stub; its separate Ollama runner is not connected to mesh jobs.

Install

bash
npm install -g vibedonate

Requires Node.js 18 or later. Includes TypeScript declarations and three direct runtime dependencies.

Run your first local check

bash
vibedonate status

On a fresh installation, the output reports that the node is not armed. No credentials or environment variables are required. No sharing starts from this command. The CLI reads exported environment variables; it does not load .env files.

For an opt-in transport demo, run this donor in one terminal:

bash
vibedonate share --compute --idle 00:00-23:59 --cap 2M --pool allowlist:consumer --handle donor

Run the consumer in another terminal:

bash
vibedonate request "hello" --pool allowlist:consumer --handle consumer

With reachable peers inside the UTC window, the consumer prints olleh. This is the echo stub, not AI inference. Stop the donor with Ctrl+C, then run vibedonate stop to revoke persisted consent.

How it works

text
share -> saved config + consent -> Hyperswarm pool discovery                                      |consumer prompt -> donor checks -> echo stub -> result + metering receipt
status / MCP preflight -> separate Ollama availability checkpayment flags -> stub wallet + local payment ledger

Announcements contain handle, pool, compute tier, and optional payment terms. Prompts and results cross connected peer links. Raw usage totals are not announced. Local files default to ~/.vibedonate.

Why it exists

vibedonate explores opt-in donation controls around spare local compute. Petals distributes model execution; Ray Serve provides model-serving infrastructure. This package focuses on consent, schedules, recipient pools, and metering, with experimental peer transport.

Compare

Checked 2026-10-14 against the Petals research paper and Ray Serve documentation.

PropertyvibedonatePetalsRay Serve
ScopeDonation controls and echo mesh demoCollaborative model inference and fine-tuningDistributed model serving
ExecutionDefault mesh uses echo stubDistributed model blocksDeployed model-serving applications
InterfaceNode.js library, CLI, agent preflightModel inference interfacePython deployment and inference APIs

Coverage

Integration or tierStatusSource behavior
HyperswarmImplementedSame-pool peer discovery and request/result transport
OllamaSeparate runnerDetects executable; invokes ollama run; not wired to mesh CLI
Echo mesh modelStubReverses prompts; records 100 tokens per default job
USDC / x402-style paymentsStubDeterministic wallet proofs and local ledger; no settlement
BaseScaffoldAccepted payment chain label
EthereumScaffoldAccepted payment chain label
PolygonScaffoldAccepted payment chain label
Subscription sharing—No shipped tier
API-credit routing—No shipped tier

There is one default chat runner, one default echo mesh model, and two MCP tools. Source inspection and local tests support these distinctions. No credentials, on-chain transactions, or public-network service guarantees were verified.

Common use cases

  1. Node.js developers: test consent and capacity decisions through the exported library.
  2. Hyperswarm experimenters: exchange echo jobs between same-pool peers.
  3. Ollama users: inspect local chat-runner availability before building a custom integration.
  4. Agent builders: call MCP request_capacity to check a proposed donation without executing a job.
  5. Vibe Suite users: inspect hash-chained usage and payment records on their own machine.

Features

  • Consent: share grants donate:compute; stop revokes it in persisted storage.
  • Schedules: UTC idle windows and configurable daily token caps.
  • Recipient pools: open, org:<id>,<member>, and allowlist:<peers>.
  • Metering: local hash-chained receipts with a chain verification method.
  • Payments: optional price metadata and stub payment records; free jobs by default.
  • Suite hooks: exported donation lifecycle hooks use vibe-core events and notifications.

Runtime surfaces

The Node.js package exports configuration parsers, consent helpers, capacity evaluation, metering, payment stubs, and compute resolution.

text
vibedonate share --compute --idle HH:MM-HH:MM --cap 2M --pool open  [--handle name] [--price 0.001] [--chain base|ethereum|polygon]vibedonate request "prompt" --pool open  [--handle name] [--timeout milliseconds] [--pay 0.001]vibedonate statusvibedonate stopvibedonate walletvibedonate mcpvibedonate --help / -hvibedonate --version / -v

vibedonate mcp starts a stdio Model Context Protocol (MCP) server with exactly two tools:

ToolBehavior
statusConfig, consent, compute availability, schedule state, and totals
request_capacityPreflight { peer, tokens }; returns an allow/deny decision without inference or a receipt

Delivery and behavior limits

The CLI donor defaults to createEchoModel(). Installing Ollama does not replace it with model inference. The separate local runtime records capacity receipts; it does not execute a prompt. The CLI uses stubWallet(). Payment success means a synthetic proof passed, not that USDC moved on Base, Ethereum, or Polygon. Peer authorization compares announced handles. It does not authenticate organization membership or bind handles to trusted identities. CLI activity checks default to “not busy”; no automatic machine-idle detector is wired into the donor. The running donor holds its initial config and consent objects. Use Ctrl+C to stop it; a separate stop process does not close it. The mesh checks whether today's total is below the cap before a job; a job can exceed the remaining allowance. The capacity preflight has a stricter request-fits check. Do not assume both paths enforce identical bounds. Discovery defaults to 15 seconds; each attempted donor gets up to six seconds. Total elapsed time can exceed the discovery timeout. Hash chains detect inconsistent record content; they do not prevent someone from rewriting the whole local ledger.

Integration guides

  • Ollama runner: install Ollama and pull qwen2.5:7b; set VIBEDONATE_OLLAMA_MODEL to choose another installed model.
  • Separate state: export VIBEDONATE_DIR to override ~/.vibedonate.
  • Demo pool (not access control): use the same allowlist:consumer string on both peers and --handle consumer for requests. Any peer that announces a matching handle passes the check.
  • Organization pool: include explicit members, such as org:team,alice; org:team alone authorizes no members.
  • Agent setup: configure your MCP host to launch vibedonate mcp over stdio.

Status

Version 0.4.3 is experimental and pre-1.0. Public case studies are not yet published. No public documentation site is configured; the operational guidance is self-contained here. Use GitHub issues for support; repository access may be required. For contributions, run npm ci, npm run typecheck, npm run build, and npm test.

FAQ

Which integrations are verified?

vibedonate has source-backed peer transport and a separate Ollama runner. Local tests do not certify public peers or blockchain settlement.

Does vibedonate provide a managed inference service?

vibedonate runs on your hardware. It provides no managed inference pool, availability guarantee, or hosted account service.

Does the CLI run Ollama for donated jobs?

vibedonate uses an echo stub for default CLI mesh jobs. The Ollama chat runner exists as a separate library path.

How is vibedonate different from Petals?

vibedonate experiments with donation controls and a request/result mesh. Petals distributes model execution across participants.

How is vibedonate different from Ray Serve?

vibedonate supplies local donation controls and a CLI experiment. Ray Serve supplies infrastructure for deployed model-serving applications.

Can I earn real USDC with vibedonate?

vibedonate ships a stub wallet and local payment records. The CLI does not submit or verify real blockchain payments.

Does stop terminate an active donor?

vibedonate stop changes persisted config and consent. Stop the active donor process with Ctrl+C to leave the mesh.

Does vibedonate share subscription quotas or API credits?

vibedonate ships only the compute tier. Subscription and API-credit routing are not implemented.

License

MIT.

来源:README.md,提交 6641b17

工具

0
SourceWeft 尚未在沙箱中启动过该软件包,因此其工具未知。

版本历史

1
  1. v0.4.2最新Sep 16, 2026

评分

已满足适用检查项的 50%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分

其他 Aibenchmarkindex 集成