AftercoreCAPACITY MARKETPLACE

AFTERCORE / GUIDES

From connection to receipt.

Instructions for the currently shipped pilot. Your model page supplies its actual price, limits and client example.

Use the reference in your editor

Markdown guides and model prices are generated from this deployment. Configuration downloads contain environment-variable placeholders, never your API key. A downloaded example is a template; running it explicitly purchases inference.

Agent-readable documentation index

Keep a client on one seat

Send an optional X-Aftercore-Session identifier (8–128 letters, digits, dots, underscores, colons or hyphens). Aftercore binds it to your API key, model and public/private scope for 30 idle minutes. Key rotation preserves the binding; another key or buyer gets a separate session.

This controls seat selection only. Send complete context on each call; it does not enable provider conversation continuation. If the pinned seat is paused, delisted, busy or otherwise ineligible, the request fails before dispatch. Wait or explicitly use a new identifier. Omit the header for normal independent selection. Each key supports up to 200 active sessions; a token-count call does not create or renew a binding.

Buy inference

  1. Open the buyer workspace. Register using the operator's pilot invitation, or sign in.
  2. Choose a model in the market. Review its protocol, price and spending hold.
  3. Open Wallet. Deposited cash and pilot credits are separate. Available balance must cover the full hold, including when a small request costs less.
  4. Create a public API key permitting the model. Its weekly budget must cover the hold as well as existing spend. Save the secret when shown; it cannot be retrieved later.
  5. Use client setup or paste the application key into the browser tester. Review the displayed price and explicitly send the request.
  6. Open Usage to see outcome, tokens, price version, supplying seat, funding and exact charge.

A public key cannot access private supply, even your own. A private key can use only your own eligible sources, under the published billing policy.

Supply capacity

  1. Open Seller dashboard and choose ChatGPT or Grok Build. Authorize the dedicated account on the provider's page using its device flow.
  2. Wait for the connected state. The credential is bound to the initiating Aftercore owner. Refresh access and models; quota stays unknown when the provider does not report it.
  3. For Grok, prepare its private seat from the connection panel. Discovered models require operator verification and a price before they can serve.
  4. Inspect the seat's public/private readiness and choose serving controls. A positive quota floor requires both windows to be reported above that floor.
  5. Explicitly list publicly. The source may receive matching demand; listing does not guarantee requests or earnings.
  6. Review served requests and funded earnings. Wallet shows payout eligibility and the status of each requested transfer.

Claude currently uses an operator-funded API credential, not a seller subscription. Its seller commission is zero.

Connect your client

Use a model's detail page or API keys → Connect your client. Choose the matching protocol and copy a Node.js, Python or bash example. Set AFTERCORE_API_KEY in your separate client environment. Do not edit the configuration of your running development agent.

OpenAI Responses, selected Chat Completions and native Anthropic Messages are implemented. Compatibility is route-specific; an example does not establish a live-tested client. Grok supports text Responses through the verified Node client. Tool calls are returned to your client; Aftercore does not execute them.

Disable automatic retries. Responses requires an Idempotency-Key; other supported routes accept one. Reusing an identity returns the recorded request instead of executing it again. Request payloads, query-string credentials and unsupported routing fields are rejected when outside the route's schema.

Billing, custody and payouts

Prices are versioned in integer nanodollars: one dollar is 1,000,000,000 nanodollars. Input, cached input and output use the recorded model tariff; Claude cache writes have separate rates. Reasoning included in output is not billed twice.

A hold reserves spending capacity before dispatch. Completed requests settle only with valid usage. Failures before output release the hold without a charge or earnings. Partial output may be billed from valid actual usage. Missing, conflicting or excessive usage stays under review and is waived after the five-minute deadline; no late automatic debit follows.

The pilot credits finalized qualifying USDG deposits on Robinhood Chain at 1 USDG = $1. Verify a dedicated wallet first; only eligible direct transfers from that wallet to the displayed custody address count. Other chains and assets are unsupported. Do not send ETH as a deposit.

Aftercore's operator controls the custody wallet and signs payouts outside the web app. The minimum is 0.000001 USDG with six-decimal transfer precision, no current withdrawal fee and operator-funded ETH gas. The UI shows the current policy. Timing depends on operator review and network finality; no processing deadline is promised. Pilot credits and unfunded accrual cannot be withdrawn. Ledger accrual does not mean a transfer was paid.

Control changes and interrupted requests

API keys shows your remaining hourly allowance for creating or rotating secrets. Revocation stays available when that allowance is exhausted. Production network limits also apply across accounts sharing an IP address. A rate-limit response includes Retry-After when a retry time is known; wait before sending again.

Delisting blocks new public requests while preserving permitted owner-private access. Pausing blocks both scopes. Revocation removes Aftercore's credential copy; revoke provider authorization separately if you want account-wide removal. Public model restrictions, quota floors and daily caps do not restrict permitted owner-private use. Concurrency covers both scopes.

Key revocation, rotation and policy changes are rechecked before dispatch. Already dispatched work may finish at its original price. Cancellation does not erase supplied output or guarantee the vendor immediately stops processing. Subscription output counts are hints rather than hard computation limits. Use Usage and private support before starting a new attempt.

Not enabled

Kimi, Antigravity, Cursor supply, open-model upstreams, Gemini generation, Claude subscription resale, token allowances/rewards and extra chains have not passed their Aftercore acceptance gates. They are retained on the roadmap. There is no automatic model substitution, paid API fallback or undisclosed subscription upgrade.