Veil

Veil is an open-source AI capacity marketplace being built in stages. Today it runs as an AI inference routing runtime for accountless, OpenAI-compatible access to Provider-supplied capacity, with Relay-based routing and signed witness recording.

Overview

Veil lets existing OpenAI-compatible clients call a local gateway while routing execution through Relay and Provider nodes. The runtime is the delivery layer for a larger product goal: Consumers get accountless access to AI capacity, Providers expose spare capacity as routable supply, Relays broker traffic without becoming the execution endpoint, and Claw becomes the low-touch automation surface for joining and operating the network.

Design Goals

• keep existing OpenAI-compatible clients working
• make AI access possible without wiring every tool directly to every upstream account
• turn spare Provider-side capacity into reusable network supply
• route around single-account or single-provider bottlenecks
• separate routing visibility from execution visibility
• preserve witness and accounting boundaries for later reconciliation
• keep crypto-compatible settlement on the main roadmap without making the runtime chain-first

Runtime Roles

• Consumer: local gateway and request origin
• Relay: verification, routing, limits, and witness
• Provider: upstream execution and streaming response delivery
• Bootstrap: Relay discovery service

The target marketplace runtime adds Claw as the operator automation surface for join, sell, pause, and recovery workflows.

Supporting Systems

• RBOB: open-source contribution accounting

Request Path

Client -> Consumer Gateway -> Relay -> Provider -> Upstream AI
                     \-> Budget      \-> Witness
Bootstrap -> Relay discovery
RBOB -> contribution accounting

What You Get

• exposes a local gateway for existing AI clients
• allows callers to consume AI access through Veil instead of binding every client directly to an upstream account
• routes encrypted requests through Relay nodes
• lets Providers contribute routable AI capacity
• executes requests on Provider nodes
• records witness data for stats, export, and later reconciliation
• keeps contribution accounting separate from production inference traffic

Product Model

• Consumer: wants reliable AI access through one local interface, without forcing every tool to manage every upstream account directly
• Provider: contributes spare AI quota or API capacity and serves inference inside the execution boundary
• Relay: brokers routing, admission, and witness without becoming the model execution host
• Claw: becomes the supported automation layer for onboarding, policy application, and sell-side operation
• Contributors: improve the protocol and runtime through the open-source build loop

Current Status

• implemented today: local Consumer gateway, Relay routing, Provider execution, witness export, wallet management, Relay discovery, and RBOB contribution accounting
• planned next: quote-aware pricing interfaces, settlement evidence contracts, payment-rail adapters, and Provider or Relay payout surfaces
• long-term target: Claw-managed network join, low-touch Provider operation, policy-driven selling, and witness-backed, crypto-compatible settlement

Quick Start

git clone https://github.com/runveil-io/core.git
cd core
npm install
npm test

Initialize a local wallet:

veil init

Run a Provider:

veil provide init
veil provide start

Run a Relay:

veil relay start

Use the local gateway:

# OpenAI-compatible endpoint
http://localhost:9960/v1

Documentation

• Docs home: docs/README.md
• 中文入口: docs/README.zh.md
• Design governance: docs/design-governance/README.md
• Product design: docs/product-design/README.md
• Trust and privacy: docs/product-design/trust-and-privacy/README.md
• Install and run: docs/installation/README.md
• Daily operations: docs/manual/README.md
• Configuration: docs/technical-design/configuration/README.md
• Client integration: docs/clients/README.md
• Architecture: docs/technical-design/architecture/README.md
• Module specs: docs/technical-design/modules/README.md

For protocol behavior, operations, and implementation boundaries, continue from docs/README.md.

Repository Layout

src/
  bootstrap/     Relay registry service
  config/        bootstrap and validation
  consumer/      local gateway and budget guard
  crypto/        signing and sealed payload handling
  discovery/     Relay discovery client
  metering/      usage normalization and pricing
  network/       WebSocket transport
  provider/      upstream execution engine
  proxy/         local secret-holding upstream proxy
  rbob/          contribution ledger
  relay/         routing, rate limit, witness
  wallet/        encrypted wallet and secrets
  cli.ts         command entry
tests/           unit and integration tests
docs/            project documentation

Runtime Defaults

• default gateway port: 9960
• default Relay port: 8080
• default Provider health port: 9962
• default proxy port: 4000
• transport: WebSocket
• runtime: Node.js 22 + TypeScript

Security Boundaries

• Relay forwards sealed request payloads without decrypting them
• Provider decrypts only inside the execution boundary
• signing keys and encryption keys are separate
• wallet files and Provider credentials are encrypted at rest
• Veil is privacy-preserving by design, but it does not claim perfect anonymity against traffic analysis, endpoint compromise, or colluding operators

Development

Read CONTRIBUTING.md and desired/*.yaml before starting work.

Useful commands:

npm test
npx vitest run --reporter=verbose
grep -rn "TODO\\|FIXME" src/

Community

Veil is an open-source project built in public. That only works if maintainers, reviewers, and contributors are visible parts of the project surface.

• Website: runveil.io
• GitHub: runveil-io
• X: @runveil_io
• Telegram: community chat
• start with CONTRIBUTING.md
• use desired/*.yaml, failing tests, and inline TODOs to find work
• open issues or pull requests when you want to improve a module, document a bug, or propose a change

Contributors

Thanks to everyone building Veil.

Contributor	PRs	Points	Contributions
@Chronolapse411	#30, #31, #33	7,500	Rate limiting, Consumer retry, Provider metrics
@sami-openlife	#15, #16	4,500	Config validation, Structured logging
@597226617	#11, #50	5,000	CLI colors and spinner, Metering module
@grit-web3-agency	#54	2,500	RBOB points ledger
@hopkdj	#12	1,500	Provider health endpoint

Points reflect the repository's contribution accounting model, including the early contributor multiplier where applicable.

License

MIT