Frequently Asked Questions
General
What is Antseed?
Antseed is a peer-to-peer network for AI services. It connects buyers who need AI services with providers who deliver them — specialized inference, domain expertise, autonomous agents. Directly, without a central platform in the middle.
Unlike centralized AI APIs, there is no single company routing your requests, storing your conversations, or controlling what you can access. The Antseed protocol handles discovery, routing, reputation scoring, and payment settlement across a decentralized network of peers.
How is this different from centralized AI chat products?
Centralized AI chat products are closed platforms built on top of a single provider's models. Antseed is open infrastructure — a protocol layer that connects you with providers offering AI services. Not just models — expertise, agents, and specialized workflows.
The key difference: your requests don't go through a single company. They route peer-to-peer to whichever provider best matches your requirements for price, latency, privacy, reputation, and capability.
Your existing tools — Claude Code, Codex, or any OpenAI-compatible client — connect and work immediately without modification.
Is Antseed open source?
Yes. The full protocol, CLI, and node implementation are open source on GitHub. Anyone can inspect the code, run a node, or build on top of the protocol.
Who runs the network?
The network runs on the independent peers that join it. Antseed is open peer-to-peer software; no single hosted service controls every node, provider, model, output, log, or data practice.
Privacy
Is my data private?
Privacy on Antseed works in two layers:
Layer 1 — Anonymous access (default). Antseed is designed for use without a central account, sign-up, or platform-issued API key. Requests route peer-to-peer, so providers generally see a peer/wallet rather than a platform identity account.
Layer 2 — TEE-verified providers. These run inside Trusted Execution Environments (Intel TDX, AMD SEV) — hardware enclaves that can reduce what the node operator can see. This is cryptographic proof, not just a policy promise. TEE provider verification is being rolled out in a future release.
Antseed's privacy model is architectural rather than account-based: users can route requests without a central identity account, platform-issued API key, or centralized chat database, and TEE providers can add hardware-backed confidentiality where available. It is not, however, a promise that every piece of data is hidden from every participant. Independent providers and supporting infrastructure may process data needed to deliver and settle requests, and public-chain activity remains visible on-chain. Users should choose providers and routes appropriate to the sensitivity of their work.
Most users are well-served by no-account routing. Journalists, lawyers, security researchers, and anyone handling sensitive data should carefully choose providers and prefer TEE routes where available.
Can providers see my prompts?
Standard providers can. TEE-verified providers cannot — the hardware prevents it even if the operator wanted to look. When paying for a TEE-verified request, you receive a cryptographic attestation proving the enclave was genuine.
Does Antseed log my requests?
Antseed the protocol does not create a central request log, but independent providers, nodes, interfaces, RPC providers, analytics tools, or other infrastructure may log or observe data. Each provider has its own data handling practices. Prefer TEE-verified providers where stronger confidentiality is required.
Can a provider see what I send?
Requests route peer-to-peer with no platform in the middle, so a provider sees a peer and a wallet, not an account. A standard provider can read the prompt it serves; a TEE-verified provider cannot, and you get an attestation to prove it. See the privacy page for a side-by-side of what each kind of provider sees.
How do I verify a provider's TEE?
Your buyer node checks it on your machine, against Intel's own certificate chain, with no third party in between and no special hardware on your side. The check happens after a provider is picked and before any payment, so attestation is free. Verification is on by default and best-effort; run antseed buyer start --verifiers antseed-verifier --require-verifier to refuse any provider that does not pass, and the request then fails with a 502 naming the failed claims instead of being quietly routed elsewhere. antseed network browse shows which providers advertise a verifier. See Verify a provider's TEE.
What does a passing TEE check prove, and what does it not?
It proves the provider node runs inside a genuine Intel TDX enclave, that the quote was minted fresh for your request and is bound to the exact peer you are paying (so replayed or borrowed quotes fail), and it can report whether the provider's GPUs run NVIDIA Confidential Computing. It proves the environment, not the model: whether the model is what the provider claims is a separate check, done by fingerprinting. See how model verification works.
Services & Availability
What services are available?
Any service a provider chooses to offer. Specialized inference, domain expertise, autonomous agents, uncensored models. The network is open — availability depends on who's online. See what's live right now on the network page.
Can I use closed-source models like GPT-4?
If a provider offers access to GPT-4 or other closed-source models, you can access them through Antseed. The protocol is model-agnostic. Availability depends on what providers choose to offer.
What if a service goes offline?
Model-only routes automatically fail over to the next eligible ranked seller after retryable peer failures. A conversation normally keeps a soft affinity to the seller that successfully served it, but the proxy can switch when that seller is unavailable, cooling down, policy-ineligible, or fails retryably. Explicit seller pins remain hard and do not switch automatically.
Setup & Usage
Do I need to sign up or create an account?
No account required. Install the CLI, run antseed buyer start, and request any model listed by curl localhost:8377/v1/models. The proxy applies the same Price + Trust preferences as the desktop AI VPN, including the default hard minimum trust score of 60. Pinning a specific seller with antseed buyer connection set --peer <peerId>, x-antseed-pin-peer, or <peerId>@<model> remains available as an explicit override. Tools that speak a supported OpenAI or Anthropic API format work immediately.
How do I install Antseed?
npm install -g @antseed/cli
Then start the buyer proxy:
antseed buyer start --router local
See the installation guide for full details.
Does it work with Claude Code, Codex, or other AI tools?
Yes. Antseed is OpenAI API-compatible. Point your tool at the Antseed endpoint and it works without any other changes — no code modifications needed.
What operating systems are supported?
macOS (Apple Silicon and Intel), Windows, and Linux. The CLI runs anywhere Node.js runs.
Are there usage limits?
No caps and no daily, weekly, or monthly quota. You pay per request from your credits and keep going as long as there are credits to spend. Top up in the AI VPN by card or Apple Pay, or with USDC from the CLI using antseed buyer deposit.
Do I need the AI VPN desktop app, or is the CLI enough?
The CLI is enough. npm install -g @antseed/cli, then antseed buyer start runs your local endpoint and antseed buyer deposit funds it with USDC, on any machine. antseed claude, antseed codex, and antseed opencode launch those tools through it. The AI VPN desktop app adds a model picker, an Apps view that launches installed tools, card and Apple Pay top-ups, and the Agents view for publishing an authenticated public endpoint. See the install guide.
Does my tool or agent need an API key?
No account and no real key. The local endpoint does not validate one, but most tools want a non-empty value, so use a placeholder like antseed. The antseed claude, codex, and opencode wrappers set this for you. Only a public endpoint you publish from the AI VPN's Agents view uses a real, generated key.
How does Antseed choose the provider for each request?
By your routing policy: price plus trust, with a default minimum trust score of 60. Leave it on auto to hit the cheapest verified provider every time, or pin a provider with <peerId>@<model> in the model field, or from the CLI with antseed buyer connection set --peer <peerId>.
Providers & Economics
Can I become a provider?
Yes. The network is open. Wrap your expertise in AI and sell it — video creation, legal analysis, code audits, uncensored inference, autonomous agents. It's your job to pick the models and workflows that deliver the best results. The network only measures delivery quality. No permission or partnership required.
See the provider setup guide to get started.
How does pricing work?
Providers set their own prices. For model-only requests, the routing layer ranks eligible offers using shared Price + Trust preferences, token or image pricing, cached-input pricing coverage, recent failures, cooldowns, and free-peer preference. The default minimum trust score is 60; explicit seller pins remain available when you need a fixed route. Prices are visible in the network-wide /v1/models catalog before routing.
How do providers get paid?
Buyers pre-deposit USDC into the on-chain AntseedDeposits contract. Each session is authorized by an EIP-712 ReserveAuth that caps the seller's charge. The seller calls reserve() on AntseedChannels to lock funds. During the session, the buyer signs cumulative SpendingAuth messages that authorize spending. The seller calls close() with the buyer's latest SpendingAuth to settle and release remaining funds. If the seller disappears, the buyer (or their deposits operator) calls requestClose(), then withdraw() after a 15-minute grace period to release remaining funds. No invoicing, no monthly billing — automatic settlement per session.
Are seller ANTS incentives claimable now?
ANTS rewards are separate from USDC payments. From epoch 22 (September 10, 2026), they depend on recognized usage, seller-pool power, and the configured reward policies. Earlier rewards use the legacy claim contracts.
Where do protocol fees go?
Network fees are set to 4% and may be directed to ecosystem mechanisms such as reserves, grants, incentives, buy-and-burn, or other community-approved uses. Any DIEM program/operator fee is governed by the separate DIEM Provider Capacity Program rules. No fee use, burn, distribution, token value, or return is promised.
How can a provider offer AI models below lab price?
Providers compete. Some run open models on their own hardware, some sell capacity they legitimately hold. Providers must build their own service on top of upstream APIs and comply with upstream terms.
Why would a provider use Antseed instead of building their own API?
Because distribution is the hard part. On Antseed a provider plugs into existing demand: buyers, discovery, reputation, and onchain settlement come with the network. No billing stack to build, no customers to acquire, no payment risk to carry. Serve a request, get paid, automatically.
Are ANTS incentives guaranteed?
No. Paying for service does not guarantee ANTS rewards. Missing pool power or a wash-trading flag can exclude usage from rewards. If you stake ANTS, withdrawing before the lock ends also incurs a principal penalty. See reward eligibility and staking terms.
The bigger picture
What happens when LLMs become so good that anyone can do anything?
That is exactly what we want. When LLMs become dramatically more capable, costs collapse and more people can run their own capable LLMs on their own hardware. Those people become Antseed providers, so the supply side grows, not shrinks. But "anyone can do anything" does not mean everyone delivers the same result. The value is in what you build on top: the skills, the workflows, the domain expertise, the agent orchestration. A more capable base model raises the ceiling for every provider.
Isn't this just like P2P file sharing? Netflix killed that.
Netflix and Spotify won because humans are happy to pay a simple subscription for a clean UI. That logic only applies to humans who care about experience. Agents don't. An agent has no preference for a polished interface, no reason to care about a brand, no inertia keeping it on a familiar platform. It just needs the service, the price, and the reliability, and on those axes an open P2P network with no middleman and no markup wins every time.