Docs

Supplier guide

Turn spare official AI capacity into USDT: what you earn, what you are responsible for, and how to connect a credential.

On this page
  1. What you earn
  2. What you are responsible for
  3. How requests reach you
  4. Add route supply
  5. Provider contract matrix
  6. Logical and upstream model mapping
  7. AWS Claude Platform credentials
  8. Amazon Bedrock Mantle credentials
  9. Protocol and streaming boundaries
  10. Before sharing
  11. Shared-route pricing
  12. Track earnings
  13. Credential safety

If you hold official AI capacity you are not fully using — promotional balances or committed spend on a model provider or cloud — you can connect it as a route. Callers use it through AnyPath, and you are paid in USDT for every request it serves.

What you earn

You receive 95% of what the caller pays; AnyPath keeps 5%. The caller pays the official list price multiplied by your provider’s multiplier.

Official list price of the usageMultiplierCaller paysYou receiveAnyPath fee
100.00 USDT0.6060.00 USDT57.00 USDT3.00 USDT

Only successful requests pay. See the current multiplier for your provider on Live supply and price.

What you are responsible for

  • Your provider account. You decide whether sharing its capacity fits the terms you accepted with the provider. If the provider limits or closes the account, that risk is yours.
  • Your quota. Requests count against your own rate limits and balances at the provider. Set RPM and TPM limits on the route if you want to keep headroom for yourself.
  • Your floor. Set a protection multiplier. While your provider’s multiplier is below it, your route does not serve, and it resumes automatically when the multiplier is raised to it or above.

You can pause or remove a route at any time.

How requests reach you

A route receives traffic only while it is eligible: the last connection check passed, the live price is at or above your floor, and it has capacity left. AnyPath chooses among eligible routes; you cannot set your own position in that order. Being eligible does not guarantee traffic.

For batch or agent-driven onboarding, use the Supply Credential API.

Add route supply

  1. Open Console -> Routes -> Add route.
  2. Choose the provider connection that matches the credential you actually own.
  3. Enter the requested API key, cloud resource metadata, one or more service-account JSON files, or delegated AWS role.
  4. Select the supported models. AnyPath sends the saved upstream identifier exactly; it does not guess or rewrite model IDs.
  5. Publish the verified route as shared supply.
  6. Save the route and run the connection check.

Do not reuse a key from a similarly named service. For example, an Anthropic Console key, an Amazon Bedrock key, and an AWS Claude Platform key are three different credentials for three different endpoints.

Provider contract matrix

The table separates the provider’s official contract from the subset currently retained by AnyPath. “Chat” means AnyPath’s OpenAI-compatible /v1/chat/completions adapter; it does not mean the provider exposes that path natively.

Provider connectionOfficial authentication and upstream endpointValue sent as modelCurrent AnyPath contract
OpenAIBearer API key at https://api.openai.com/v1OpenAI model IDChat, Responses, image generation, image edits
Anthropic Consolex-api-key and anthropic-version at https://api.anthropic.comAnthropic model IDMessages, token counting, translated Chat
Gemini API / AI Studiox-goog-api-key at https://generativelanguage.googleapis.com/v1betaGemini model IDNative Gemini /v1beta, translated Chat, image generation
Gemini on Vertex AIOAuth bearer token derived from one or more service-account JSON files at the selected project/locationVertex Gemini model IDNative Gemini /v1beta
Claude on Vertex AIOAuth bearer token derived from one or more service-account JSON files; regional Vertex publisher endpointVersioned Vertex Claude ID, such as claude-...@YYYYMMDDMessages through the native Vertex route
AWS Claude PlatformAWS Claude Platform workspace API key as Bearer plus anthropic-workspace-id; https://aws-external-anthropic.<region>.api.awsClaude model IDMessages, token counting, translated Chat
Amazon Bedrock Mantle ClaudeMantle API key, or SigV4 with STS credentials obtained from a delegated IAM role; https://bedrock-mantle.<region>.api.aws/anthropic/v1/messagesExact in-region anthropic.* model IDNative Messages and token counting
xAIBearer API key at https://api.x.ai/v1xAI model IDChat, Responses, image generation; image edits are not retained
OpenRouterBearer API key at https://openrouter.ai/api/v1Exact curated provider/model slugChat and stateless Responses; other OpenRouter API families are not retained
Vercel AI GatewayAI Gateway key at https://ai-gateway.vercel.sh/v1; Bearer for OpenAI-compatible paths and x-api-key for Anthropic MessagesExact provider-qualified gateway model IDChat, Responses, Messages, token counting, and image generation

Every retained provider connection can be shared. Shared-route pricing uses the model’s official list price multiplied by the actual routed Provider’s platform multiplier and metered usage. The official list price used for charging and the provider that actually served the request may differ, and both remain part of the settlement evidence. For aggregate gateways, AnyPath only exposes curated, provider-qualified model mappings, pins the serving provider, and binds settlement to that model’s official list price. Dynamic selectors such as openrouter/auto are not offered because they cannot fix an official list price before routing. OpenRouter’s Responses API is stateless: do not depend on stateful continuation fields such as previous_response_id.

For Claude on Vertex AI, AnyPath intentionally uses the matching Anthropic direct official model price for settlement. This is a user-approved AnyPath platform pricing assumption, not a statement about the price shown on the Google Cloud or Vertex invoice. Gemini on Vertex AI uses the matching Google official list price. A Vertex route accepts one JSON object, a JSON array, or multiple uploaded service-account JSON files; enabled accounts are retained as distinct credential slots for deterministic rotation and failover.

Only the curated models and protocol families shown by Console are shareable; upstream support alone does not make an arbitrary model or mode routeable. Anthropic Messages token counting continues to use POST /v1/messages/count_tokens with the Messages authentication contract.

Official references:

Logical and upstream model mapping

The model ID used by your customers does not need to match the provider’s model ID. Save both identities in the channel’s model mapping: for example, logical claude-sonnet-4-5 maps to Vertex claude-sonnet-4-5@20250929. AnyPath keeps the logical identity for routing, billing and protocol-defined response model fields, and sends the saved upstream identity in provider URLs or request bodies. Changing a mapping requires a new connection check; an old successful probe cannot certify the new identity.

The Console’s provider catalog supplies curated defaults. Per-channel mappings are the runtime authority. A default mapping or a successful connection check alone does not mean shared capacity is currently available; account readiness, operator review, usage evidence and capacity limits still apply.

AWS Claude Platform credentials

Select the AWS commercial Region where the workspace was created and enter its exact wrkspc_... ID. Region is required, and a workspace can only be reached through the endpoint for its own Region. Opt-in AWS Regions must already be enabled for the owning AWS account. AnyPath rejects GovCloud, China, unknown Region codes, and workspace IDs that do not use the documented format.

Amazon Bedrock Mantle credentials

For a durable production connection, choose AWS AssumeRole and enter the role ARN generated for your AWS account. AnyPath supplies account-scoped trust material and a restricted Bedrock permission policy. Before saving the route, it verifies all of the following through AWS STS:

  • the role trusts the exact AnyPath AWS principal;
  • the trust policy requires AnyPath’s generated External ID;
  • the role cannot be assumed with the External ID omitted or changed;
  • the role rejects a missing or changed SourceIdentity and accepts the generated value.

The generated role policy contains only bedrock-mantle:CreateInference, bedrock-mantle:ListModels, and bedrock-mantle:GetModel for the account project. The later connection check confirms that those Mantle permissions work; the STS trust check itself does not inspect the attached permission policy.

AnyPath calls AWS STS AssumeRole, keeps the returned temporary credentials only in the runtime credential cache, and refreshes them before expiry. The Console does not accept or store an IAM access key, secret access key, or manually pasted session token. This follows AWS’s third-party role guidance and AssumeRole contract.

A Mantle API key remains available as a separate connection mode through the native x-api-key header. AWS recommends long-term Bedrock API keys only for exploration; production applications should use temporary credentials. See How Bedrock API keys work.

AnyPath uses only Bedrock Mantle. It does not call Bedrock Runtime, accept cross-region inference-profile IDs, or keep a Runtime fallback. Mantle receives the native Anthropic Messages body unchanged, including the exact in-region anthropic.* model ID, and returns native JSON or SSE. The connection probe validates the selected model with a real Mantle request; the account’s GET /v1/models result remains the authoritative discovery view.

AWS currently documents Mantle endpoints in 13 commercial Regions plus us-gov-west-1. Claude availability is narrower: after intersecting Mantle endpoint Regions, model deployment Regions, AnyPath’s commercial AWS role contract, and the retained model catalog, the supported model-region combinations are:

  • us-east-1: Claude Sonnet 5, Claude Haiku 4.5, Claude Opus 4.7, and Claude Opus 4.8;
  • eu-north-1 and eu-west-1: Claude Haiku 4.5, Claude Opus 4.7, and Claude Opus 4.8;
  • ap-northeast-1: Claude Haiku 4.5 and Claude Opus 4.8.

AnyPath currently enables Bedrock Mantle shared supply in us-east-1 only, because shared-route settlement requires one explicit region-specific official list price.

AWS’s Mantle compatibility table lists seven Claude IDs: anthropic.claude-sonnet-5, anthropic.claude-mythos-5, anthropic.claude-fable-5, anthropic.claude-mythos-preview, anthropic.claude-haiku-4-5, anthropic.claude-opus-4-7, and anthropic.claude-opus-4-8. AnyPath enables the four general models that do not require provider data sharing by default. Mythos 5 and Fable 5 require provider_data_share; Mythos 5 and Mythos Preview require gated access. Those three models remain non-routeable by default and should be enabled only after the applicable access, privacy, and pricing review plus an account-level discovery check.

Protocol and streaming boundaries

  • Only the contract families listed above are routeable. An API supported by the upstream provider is not automatically part of AnyPath’s public gateway contract.
  • Cross-protocol adapters are used only where listed. If headers, body shape, usage evidence, or streaming semantics cannot be preserved, the route fails closed.
  • Streaming works on channels that have passed streaming certification: the connection probe makes a real streaming request and verifies terminal SSE usage evidence (sse_terminal in channel_model_usage_evidence). Streaming requests to uncertified channels fail closed with model_unavailable. Certified streams settle from that terminal usage evidence; capture never exceeds the pre-authorized hold, and a stream that ends without terminal usage releases the hold.
  • Token counting requires its own successful endpoint probe and request-unit evidence. A successful generation request does not certify token counting; Claude on Vertex AI does not expose it through AnyPath. Probe evidence alone does not make a count-only route shareable: its canonical price must support request-unit billing, so a per-token model price does not qualify.
  • Model availability is the intersection of the provider connection, the saved model mapping, current health, and AnyPath’s public contract family.

Before sharing

A shared route should be published only after:

  • credentials are saved and encrypted;
  • the connection test succeeds;
  • every logical model has an exact upstream model mapping;
  • every model has an official list price in the catalog;
  • RPM, TPM, budget, and payout settings are ready; and
  • the route can produce terminal usage evidence required for settlement.

Shared-route pricing

For successful shared usage:

  • the caller pays the model’s official list price multiplied by your provider’s multiplier and the metered usage;
  • you receive 95% of that amount;
  • AnyPath keeps a 5% service fee;
  • if the caller was referred, the referrer receives an extra 2%, paid out of AnyPath’s fee; and
  • failed requests do not settle.

If you recommend a customer, their shared requests prefer your healthy supply when it matches their route policy. If your supply cannot serve that request, AnyPath continues with other eligible shared routes without increasing the customer’s price.

Track earnings

Use Console -> Billing -> Earnings & payouts to review gross usage, AnyPath fees, net earnings, referral rewards, payout readiness, and treasury-rail status.

Credential safety

Provider API keys and service-account JSON are encrypted after save and are not shown again. Delegated AWS role metadata is stored without temporary STS credentials. Never put a provider secret in a route name, model mapping, support ticket, or request log.