12 focused AI services · USDC on Base

Prepare your data.
Run your AI. Measure its output.

Small, clearly defined jobs your software can buy through an API. Dataset preparation, answer comparison, local inference, search tools and a downloadable small-model training adapter.

Prices below cover one job within its published limits. These are bounded services on one PC; no enterprise throughput or availability guarantee is offered.

Browse services · Connect your agent · Data handling · Machine-readable collection

Connect and pay

POST https://market.datapackvibe.com/x402/ai-text-summary
Content-Type: application/json

{"text":"Your text to summarize..."}

The first request returns HTTP 402 and the price. An x402 client authorizes USDC on Base and retries. The service completes before payment is settled; failed computation is not charged. Results include downloadable files. Existing Agent Market wallets can also use these SKUs through market_buy or /v1/buy.

Allow up to 90 seconds for a model job. Busy workers reject fresh jobs without settlement; retry later. The original AI collection accepts JSON text and records directly, with no arbitrary model loading or customer code execution. Separate Assistant services include a public static webpage reader, DNS email-domain checks and public token market snapshots; their catalogue schemas describe the limits.

Services and limits

Training Dataset Cleanup

0.02 USDC per completed job

Trim outer whitespace, reject empty records, remove exact duplicates and export JSONL. Up to 500 flat records; no semantic deduplication.

POST /x402/ai-dataset-clean

Request example and input schema

{
  "records": [
    {
      "text": "Delivery was quick.",
      "label": "positive"
    },
    {
      "text": "The package arrived damaged.",
      "label": "negative"
    },
    {
      "text": "Delivery was quick.",
      "label": "positive"
    }
  ]
}

{
  "type": "object",
  "properties": {
    "records": {
      "type": "array",
      "items": {
        "type": "object",
        "additionalProperties": {
          "type": [
            "string",
            "number",
            "boolean",
            "null"
          ]
        }
      },
      "minItems": 1,
      "maxItems": 500,
      "default": [
        {
          "text": "Delivery was quick.",
          "label": "positive"
        },
        {
          "text": "The package arrived damaged.",
          "label": "negative"
        },
        {
          "text": "Delivery was quick.",
          "label": "positive"
        }
      ]
    }
  },
  "required": [
    "records"
  ],
  "additionalProperties": false
}

Training Dataset Quality Report

0.03 USDC per completed job

Check missing fields/labels, label imbalance, duplicate texts, conflicting labels and train/validation overlap. Up to 500 training and 500 validation records.

POST /x402/ai-dataset-quality

Request example and input schema

{
  "training": [
    {
      "text": "Delivery was quick.",
      "label": "positive"
    },
    {
      "text": "The package arrived damaged.",
      "label": "negative"
    },
    {
      "text": "Delivery was quick.",
      "label": "positive"
    }
  ],
  "validation": [
    {
      "text": "Delivery was quick.",
      "label": "positive"
    }
  ],
  "text_field": "text",
  "label_field": "label"
}

{
  "type": "object",
  "properties": {
    "training": {
      "type": "array",
      "items": {
        "type": "object",
        "additionalProperties": {
          "type": [
            "string",
            "number",
            "boolean",
            "null"
          ]
        }
      },
      "minItems": 1,
      "maxItems": 500,
      "default": [
        {
          "text": "Delivery was quick.",
          "label": "positive"
        },
        {
          "text": "The package arrived damaged.",
          "label": "negative"
        },
        {
          "text": "Delivery was quick.",
          "label": "positive"
        }
      ]
    },
    "validation": {
      "type": "array",
      "items": {
        "type": "object",
        "additionalProperties": {
          "type": [
            "string",
            "number",
            "boolean",
            "null"
          ]
        }
      },
      "maxItems": 500,
      "default": [
        {
          "text": "Delivery was quick.",
          "label": "positive"
        }
      ]
    },
    "text_field": {
      "type": "string",
      "maxLength": 40,
      "minLength": 1,
      "default": "text"
    },
    "label_field": {
      "type": "string",
      "maxLength": 40,
      "minLength": 1,
      "default": "label"
    }
  },
  "required": [
    "training",
    "validation",
    "text_field",
    "label_field"
  ],
  "additionalProperties": false
}

AI Answer Evaluation and Comparison

0.03 USDC per completed job

Score two sets of answers against customer references with normalized exact match and token F1. Up to 200 examples. These metrics do not verify factual truth.

POST /x402/ai-answer-evaluation

Request example and input schema

{
  "examples": [
    {
      "reference": "Paris",
      "answer_a": "Paris",
      "answer_b": "London"
    }
  ]
}

{
  "type": "object",
  "properties": {
    "examples": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "reference": {
            "type": "string",
            "maxLength": 1500,
            "minLength": 1
          },
          "answer_a": {
            "type": "string",
            "maxLength": 1500,
            "minLength": 1
          },
          "answer_b": {
            "type": "string",
            "maxLength": 1500,
            "minLength": 1
          }
        },
        "required": [
          "reference",
          "answer_a",
          "answer_b"
        ],
        "additionalProperties": false
      },
      "minItems": 1,
      "maxItems": 200,
      "default": [
        {
          "reference": "Paris",
          "answer_a": "Paris",
          "answer_b": "London"
        }
      ]
    }
  },
  "required": [
    "examples"
  ],
  "additionalProperties": false
}

AI Structured Output Validation

0.01 USDC per completed job

Validate a JSON value against a documented schema subset: types, properties, required, items, enums and size/number limits. No external references or regex execution.

POST /x402/ai-output-validation

Request example and input schema

{
  "value": {
    "name": "Ada",
    "count": 2
  },
  "schema": {
    "type": "object",
    "properties": {
      "name": {
        "type": "string"
      },
      "count": {
        "type": "integer",
        "minimum": 0
      }
    },
    "required": [
      "name",
      "count"
    ],
    "additionalProperties": false
  }
}

{
  "type": "object",
  "properties": {
    "value": {
      "default": {
        "name": "Ada",
        "count": 2
      }
    },
    "schema": {
      "type": "object",
      "default": {
        "type": "object",
        "properties": {
          "name": {
            "type": "string"
          },
          "count": {
            "type": "integer",
            "minimum": 0
          }
        },
        "required": [
          "name",
          "count"
        ],
        "additionalProperties": false
      }
    }
  },
  "required": [
    "value",
    "schema"
  ],
  "additionalProperties": false
}

AI Text Summary

0.02 USDC per completed job

Summarize up to 1500 characters in a short paragraph. Local Qwen3 small-model inference; output requires review.

POST /x402/ai-text-summary

Request example and input schema

{
  "text": "Northwind support opens Monday to Friday at 09:00 and closes at 17:00. Customers can request a refund within thirty days."
}

{
  "type": "object",
  "properties": {
    "text": {
      "type": "string",
      "maxLength": 1500,
      "minLength": 1,
      "default": "Northwind support opens Monday to Friday at 09:00 and closes at 17:00. Customers can request a refund within thirty days."
    }
  },
  "required": [
    "text"
  ],
  "additionalProperties": false
}

AI Text Classification

0.02 USDC per completed job

Classify up to 1200 characters into one of 2..6 customer labels. Returns a label and brief rationale; no calibrated confidence claim.

POST /x402/ai-text-classification

Request example and input schema

{
  "text": "Please refund the damaged order.",
  "labels": [
    "billing",
    "delivery",
    "technical"
  ]
}

{
  "type": "object",
  "properties": {
    "text": {
      "type": "string",
      "maxLength": 1200,
      "minLength": 1,
      "default": "Please refund the damaged order."
    },
    "labels": {
      "type": "array",
      "items": {
        "type": "string",
        "maxLength": 40,
        "minLength": 1
      },
      "minItems": 2,
      "maxItems": 6,
      "default": [
        "billing",
        "delivery",
        "technical"
      ]
    }
  },
  "required": [
    "text",
    "labels"
  ],
  "additionalProperties": false
}

AI Structured Field Extraction

0.03 USDC per completed job

Extract up to five named fields from up to 1200 characters. Each value must be an exact source substring or null, with missing information left empty.

POST /x402/ai-field-extraction

Request example and input schema

{
  "text": "Customer: Ada Lovelace. Order: AB123. Status: shipped.",
  "fields": [
    "customer",
    "order",
    "status"
  ]
}

{
  "type": "object",
  "properties": {
    "text": {
      "type": "string",
      "maxLength": 1200,
      "minLength": 1,
      "default": "Customer: Ada Lovelace. Order: AB123. Status: shipped."
    },
    "fields": {
      "type": "array",
      "items": {
        "type": "string",
        "maxLength": 40,
        "minLength": 1
      },
      "minItems": 1,
      "maxItems": 5,
      "default": [
        "customer",
        "order",
        "status"
      ]
    }
  },
  "required": [
    "text",
    "fields"
  ],
  "additionalProperties": false
}

AI Text Embeddings

0.03 USDC per completed job

Generate normalized 384-dimensional MiniLM embeddings for up to 32 texts, each up to 1000 characters. Token truncation is reported.

POST /x402/ai-text-embeddings

Request example and input schema

{
  "texts": [
    "How do I request a refund?",
    "Refunds are available within thirty days."
  ]
}

{
  "type": "object",
  "properties": {
    "texts": {
      "type": "array",
      "items": {
        "type": "string",
        "maxLength": 1000,
        "minLength": 1
      },
      "minItems": 1,
      "maxItems": 32,
      "default": [
        "How do I request a refund?",
        "Refunds are available within thirty days."
      ]
    }
  },
  "required": [
    "texts"
  ],
  "additionalProperties": false
}

AI Document Search Preparation

0.05 USDC per completed job

Split up to 12000 characters of supplied plain text into overlapping chunks and produce MiniLM embeddings. Returns a portable search dataset; not a hosted database.

POST /x402/ai-document-preparation

Request example and input schema

{
  "text": "Northwind support opens Monday to Friday at 09:00 and closes at 17:00. Customers can request a refund within thirty days."
}

{
  "type": "object",
  "properties": {
    "text": {
      "type": "string",
      "maxLength": 12000,
      "minLength": 1,
      "default": "Northwind support opens Monday to Friday at 09:00 and closes at 17:00. Customers can request a refund within thirty days."
    }
  },
  "required": [
    "text"
  ],
  "additionalProperties": false
}

AI Search Result Reranking

0.03 USDC per completed job

Rank up to 20 passages by relevance to a query using a TinyBERT cross-encoder. Returns original indices and uncalibrated scores, with truncation disclosed.

POST /x402/ai-search-reranking

Request example and input schema

{
  "query": "How long do I have to request a refund?",
  "documents": [
    "Refunds are available within thirty days.",
    "The office opens at nine."
  ]
}

{
  "type": "object",
  "properties": {
    "query": {
      "type": "string",
      "maxLength": 400,
      "minLength": 1,
      "default": "How long do I have to request a refund?"
    },
    "documents": {
      "type": "array",
      "items": {
        "type": "string",
        "maxLength": 1500,
        "minLength": 1
      },
      "minItems": 1,
      "maxItems": 20,
      "default": [
        "Refunds are available within thirty days.",
        "The office opens at nine."
      ]
    }
  },
  "required": [
    "query",
    "documents"
  ],
  "additionalProperties": false
}

AI Synthetic Training Examples

0.04 USDC per completed job

Generate three short prompt/completion drafts from a task description and supplied reference. Local small-model output, explicitly unverified and requiring review.

POST /x402/ai-synthetic-examples

Request example and input schema

{
  "task": "Draft customer questions and short answers about the refund policy.",
  "reference": "Northwind support opens Monday to Friday at 09:00 and closes at 17:00. Customers can request a refund within thirty days."
}

{
  "type": "object",
  "properties": {
    "task": {
      "type": "string",
      "maxLength": 300,
      "minLength": 1,
      "default": "Draft customer questions and short answers about the refund policy."
    },
    "reference": {
      "type": "string",
      "maxLength": 1000,
      "minLength": 1,
      "default": "Northwind support opens Monday to Friday at 09:00 and closes at 17:00. Customers can request a refund within thirty days."
    }
  },
  "required": [
    "task",
    "reference"
  ],
  "additionalProperties": false
}

Small Language Model LoRA Fine-Tuning

0.25 USDC per completed job

Train a downloadable SmolLM2-135M-Instruct LoRA adapter on 2..16 short prompt/completion examples. One epoch, 128 tokens per example including template. Small pilot job; no quality guarantee.

POST /x402/ai-small-model-finetune

Request example and input schema

{
  "examples": [
    {
      "prompt": "How can I request a refund?",
      "completion": "Contact support within thirty days of purchase."
    },
    {
      "prompt": "When is support open?",
      "completion": "Monday to Friday, from 09:00 to 17:00."
    }
  ]
}

{
  "type": "object",
  "properties": {
    "examples": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "prompt": {
            "type": "string",
            "maxLength": 200,
            "minLength": 1
          },
          "completion": {
            "type": "string",
            "maxLength": 200,
            "minLength": 1
          }
        },
        "required": [
          "prompt",
          "completion"
        ],
        "additionalProperties": false
      },
      "minItems": 2,
      "maxItems": 16,
      "default": [
        {
          "prompt": "How can I request a refund?",
          "completion": "Contact support within thirty days of purchase."
        },
        {
          "prompt": "When is support open?",
          "completion": "Monday to Friday, from 09:00 to 17:00."
        }
      ]
    }
  },
  "required": [
    "examples"
  ],
  "additionalProperties": false
}

Fine-tuning: a small pilot with real weights

The training service updates LoRA adapter weights for SmolLM2-135M-Instruct, a small language model. It accepts 2–16 examples, trains one epoch and returns a ZIP containing safetensors adapter weights, configuration, loss measurements and loading instructions. Each example must fit 128 model tokens, including its chat template.

This is a bounded experiment. It does not train a large model from scratch, provide ongoing model hosting or establish that a tuned model is better. Training loss is reported separately from held-out evaluation; none is claimed. Use separate representative examples to assess the adapter.

How to interpret results

Dataset cleanup removes exact duplicates after trimming outer string whitespace. Quality overlap detection normalizes Unicode, letter case and whitespace.
Answer evaluation measures agreement with your reference answers, using exact match and token overlap F1. It does not independently establish factual correctness.
Embeddings use MiniLM with mean pooling and unit normalization. Inputs beyond 256 model tokens are truncated and flagged. Document preparation returns chunk text, offsets and embeddings.
Reranking uses a TinyBERT cross-encoder with 512-token pairs. Scores rank relevance; they are not probabilities. Truncated pairs are flagged.
Model-generated summaries, labels, extractions and synthetic examples require review. Synthetic examples are explicitly marked as drafts.
Output validation supports a JSON Schema subset: type, properties, required, additionalProperties (boolean), items, scalar enum, minLength/maxLength, minimum/maximum and minItems/maxItems. Unsupported keywords are rejected.

Data handling

Local AI services process content on the seller's PC. Services explicitly labelled hosted send the supplied prompts, queries or text to BlockRun and its named model or data provider. Hosted-service descriptions identify those providers and limits before purchase. Supplier responses and generated media metadata are retained for delivery and payment reconciliation. Assistant webpage readers send the requested URL to the public website and follow up to two validated redirects; DNS checks disclose the email domain to DNS resolvers; token snapshots send the token address to DEX Screener. Inputs are passed in memory and are not intentionally logged or saved as input files. Temporary training workspaces are removed after completion or failure.

Delivered outputs and order results are retained in the shop's storage and may be included in backups. Download links expire after seven days; link expiry does not delete the stored content. Automatic deletion is not currently promised. Derived outputs, embeddings and adapters can contain or reveal sensitive information.

Only an explicitly purchased fine-tuning job uses your examples to train your adapter. Other requests are not reused for training. Payment processing uses Coinbase; the public Bazaar listing contains synthetic example requests and service metadata, not your private job content.

Prepare your data.Run your AI. Measure its output.

Connect and pay

Services and limits

Training Dataset Cleanup

Training Dataset Quality Report

AI Answer Evaluation and Comparison

AI Structured Output Validation

AI Text Summary

AI Text Classification

AI Structured Field Extraction

AI Text Embeddings

AI Document Search Preparation

AI Search Result Reranking

AI Synthetic Training Examples

Small Language Model LoRA Fine-Tuning

Fine-tuning: a small pilot with real weights

How to interpret results

Data handling

Prepare your data.
Run your AI. Measure its output.