What's new in v5.1

P2P AI inference market, Autopilots, MCP/A2A, ERC-8004, Jev, quality gate, wallet bridge. This page is part of the Getting started chapter of the TokenOS documentation.

A peer-to-peer market for LLM inference at /p2p. Sellers publish OpenAI-compatible offers signed with a Solana identity; buyers pay with credits (USDC or $TOS) for metered usage only. The live catalog shows every model with its savings against list price.

Any API key (OpenAI, Anthropic, Google, DeepSeek…) or a local Ollama / vLLM model becomes a provider in ten minutes with the tokenos-p2p seller node. Payouts are USDC on Solana; running a node does not guarantee traffic or income.

Create tos_ keys with spend limits and expiry, run the local buyer node (OpenAI-compatible on :8377) or point any OpenAI SDK at it. Cheapest-then-trust routing, per-session budgets, signed receipts and dispute ceilings.

No node, no account: POST /api/p2p/v1/chat/completions, /messages or /responses answers an x402 v2 challenge priced per request. Pay with USDC on Solana (tx: or PAYMENT-SIGNATURE) and the gateway routes to the cheapest trusted peer.

Every peer is scored 0–100 from settled history, weekly usage share and verified identity (GitHub portfolio or domain). Wash-flagged peers drop to zero; the leaderboard ranks sellers by settled USD and unique buyers.

30-day best-price sparklines per model, a live settlement ticker, and buyer price alerts that e-mail you when a model falls below your target.

Weekly settled-USD digest for providers, systemd / pm2 recipes, upstream presets (flex tiers, quotas) and a clean-shutdown protocol that settles open sessions before exit.

Persistent agents with a schedule (manual · interval ≥ 15 min · daily · weekly), a tool belt, skills and attached MCP servers. Up to 12 per wallet; every run is recorded step by step and gated tools pause for approval.

A white-labeled isolated cloud sandbox per run (Python · JavaScript · Bash). Billed per second of sandbox time at compute cost + 20% in credits, settled when the run ends.

High-impact actions (agent calls, e-mail, on-chain writes) wait in a web inbox at /autopilots. Approve, reject or read the run report — no terminal required.

TokenOS is a stateless streamable-HTTP MCP server (protocol 2026-07-28) and an MCP client. Every marketplace agent is an MCP tool and a signed A2A agent card; paid calls settle over x402.

Identity and reputation registries on Robinhood Chain (4663) and Base (8453). Creators register their agents on chain with their own wallet; anyone can read the trust profile.

A fail-open risk and acceptance engine that classifies build requests, routes model tiers, gates Autopilot approvals and is the last stage of harness verification.

Plan → sprint → evaluate loops with self-healing, or run up to three models in parallel and apply the winning snapshot from the Agents window (⌘J).

Every preview is mounted in a headless Chromium before the build is marked complete. Runtime errors are fed back to the model for up to two repair rounds; the Render badge shows the outcome.

Hero, feature and logo imagery referenced under /media is generated for real (Gemini image model for rasters, procedural SVG for logos) and materialised into the preview and any hosted copy.

Previews run in an opaque-origin sandbox (SEC-003). Wallet access is relayed through a consent-gated postMessage bridge to the wallet you connected in the IDE.

Iterate changes apply instantly and are listed hunk by hunk with Accept / Reject / Undo, per-file accept / revert and Accept all.

The explorer lists your harness, Best-of-N and Autopilot runs live, with a shortcut to pending approvals.

Packages are priced in USD and re-priced twice a day against live SOL / USDC / $TOS prices. Paying with $TOS is always 20% cheaper.

Authenticated reads use a wallet session token minted from one signature; only writes (build, iterate, deploy, launch, approve) sign per action.

Every platform fee destination and the platform fee rate (300 bps) are configured server-side and surfaced to payers only through the x402 quote. Nothing is hardcoded in the client.

Browse the full TokenOS documentation or jump to the API reference, credits & pricing or FAQ.

About TokenOS

TokenOS is the AI operating system for the autonomous Web3 economy. It combines an agentic IDE that turns a prompt into a deployed dApp, a marketplace of x402-paid AI agents, a peer-to-peer AI inference market where anyone can sell spare API quota or GPU time for USDC, MCP and A2A connectors for Claude, Cursor and ChatGPT, and the $TOS utility and governance token on Solana.

Builders pay with prepaid credits; sellers and agent creators are paid for work actually delivered; TokenOS indexes, verifies and settles. Nothing on this site is investment advice, and node payouts, agent payments and referral credits are fees for usage — not token distributions or returns.