Somnia LLM Inference

Somnia LLM Inference runs large language models inside Somnia, powered by the Somnia Validator Network. Every validator in a subcommittee runs the same pinned model on the same input and must return a byte-identical answer, so the result a contract receives is consensus-validated, verifiable and free of single-party control.

Deterministic AI your contracts can settle on.

Why Somnia LLM Inference

Same prompt, same answer, on every validator

Fixed seeds, zero temperature and a pinned open-weight model make inference deterministic, so independent validators reproduce the exact same output.

Consensus instead of a single provider

A single model endpoint can be manipulated or switched off. Answers on Somnia are agreed by a subcommittee of validators before a contract sees them.

Inspect the reasoning

Every execution produces signed receipts of the steps each validator took, including optional chain-of-thought, so you can audit how an answer was reached.

AI as a primitive, not a dependency

Contracts call the model like any other function and get the result in a callback. No API keys, no offchain service to run and no trusted middleman.

How an inference request runs

Request. A contract sends a prompt to the Somnia Agents platform with a small deposit and a callback.
Elect. A subcommittee of validators from the Somnia Validator Network is chosen to run the request.
Infer. Each validator runs the same pinned model with a fixed seed and zero temperature.
Agree. Results are compared. Under majority consensus, byte-identical answers from the threshold of validators finalise the request.
Deliver. The agreed answer is returned to the contract's callback, with signed receipts available for audit and unspent deposit refunded.

Capabilities

inferString. Single-turn text, optionally constrained to a list of allowed values.
inferNumber. An integer answer clamped to the range you set.
inferChat. Multi-turn conversations with system, user and assistant roles.
inferToolsChat. Tool use with MCP servers and onchain functions returned as calldata.
Chain-of-thought. Optional reasoning, recorded in the execution receipts.
Majority or threshold consensus. Accept identical answers, or receive every result and aggregate onchain.

Example

moderation.ts

// From docs.somnia.network/agents/base-agents/llm-inference
const abi = [{
  type: 'function',
  name: 'inferString',
  inputs: [
    { name: 'prompt', type: 'string' },
    { name: 'system', type: 'string' },
    { name: 'chainOfThought', type: 'bool' },
    { name: 'allowedValues', type: 'string[]' }
  ],
  outputs: [{ name: 'response', type: 'string' }]
}];

// Content moderation example — constrain output to "safe" or "unsafe"
const calldata = encodeFunctionData({
  abi,
  functionName: 'inferString',
  args: [
    'Analyze the following text: "Check out this amazing new product!"',
    'You are a content moderation assistant.',
    false,            // chainOfThought
    ['safe', 'unsafe'] // allowedValues
  ]
});