# NeuroDynamic LLM API

Pay per request with USDC on Base. No NanoGPT account or subscription required.

Models and current prices: https://api.neurodynamic.tech/v1/llm/models
API schema: https://api.neurodynamic.tech/v1/llm/openapi.json
Terms: https://api.neurodynamic.tech/v1/llm/terms
Privacy: https://api.neurodynamic.tech/v1/llm/privacy
Client: https://api.neurodynamic.tech/v1/llm/client.py

| Model ID | Price per request | Intended use |
| --- | --- | --- |
| abliterated | 0.020 USDC | Reduced-refusal writing, chat and general text |
| abliterated-large | 0.035 USDC | Larger reduced-refusal model for writing and reasoning tasks |

Both are uncensored model variants from Abliteration.ai accessed through NanoGPT. Lawful-use restrictions still apply. This is an introductory text-only beta, not a complete OpenAI-compatible API. No streaming, tools, file input, image input or conversation storage. Maximum input: 2000 UTF-8 bytes. Output: 16–512 tokens, default 512. A finish_reason of length means the answer stopped at your token limit.

## Request flow

1. Discover models and read the terms/privacy notice. GET requests are free.
2. POST to /v1/llm/{model}/completions without a payment header for a free HTTP402 quote. This does not call an LLM or spend money.
3. Check the resource URL, model price, Base network and USDC recipient. Obtain the user's spending authority before signing. Save the exact request and payment proof privately on the requester side.
4. POST the same endpoint with JSON and PAYMENT-SIGNATURE. We verify authorisation, generate text, settle the quoted amount and return text plus PAYMENT-RESPONSE.
5. On uncertainty, preserve the proof. POST /v1/llm/{model}/status with that proof to check payment status. Do not automatically sign again.

Example JSON:

```json
{"input":"Write a friendly one-sentence welcome.","max_tokens":128,"terms_version":"2026-09-11-llm-v1","adult":true,"third_party_processing":true}
```

The adult and processing acknowledgements must reflect the actual requester; agents must not invent consent. Terms are versioned so that a client cannot silently accept a future material change.

Successful JSON contains model, text, finish_reason, token usage, ai_generated=true and request_id. The response header PAYMENT-RESPONSE carries the x402 receipt. 503/429 before dispatch means retry later using the same proof; 409 means check status and contact support; 410 means this proof already settled and the output is not retained. The service never automatically generates another answer using the same proof.

**Content privacy:** prompts and answers are held only in memory while processing on NeuroDynamic servers. No local content history or recovery cache. Metadata and payment receipts are retained separately. Cloudflare, NanoGPT and the named model provider process requests under their own policies; our promise does not extend to their servers. A lost paid answer cannot be resent: contact support@neurodynamic.tech with the request ID for resolution.

The downloadable client defaults to a free quote. Explicit --pay is required, and it saves proof before sending. It stores your own input and result locally on your machine; that is separate from the server's no-content-retention policy. Keep its files private.