Liminal (Model) Adapters beta

Focus a model adapter with L.focus. The specified client adapter will be used for subsequent inference.

AI SDK Integration

import { openai } from "@ai-sdk/openai"
import { L } from "liminal"
import { ai } from "liminal-ai"

function* g() {
  yield* L.focus(ai(openai("gpt-4o-mini")))
}

Ollama Integration

import { L } from "liminal"
import { ollama } from "liminal-ollama"

function* g() {
  yield* L.focus(ollama("gemma3:1b"))
}

Switch Models

import { adapter } from "liminal-ollama"

function* g() {
  yield* L.focus(adapter("gpt-4o-mini"))
  yield* L.assistant // uses 4o mini
  yield* L.focus(adapter("o4-mini-high"))
  yield* L.assistant // uses 4o mini high
}

Creating Adapters

import { Adapter } from "liminal"
import { type Message, Ollama } from "ollama"

// Up to you. Could be a union of model name literals.
type YourConfig = any

export function adapter(config: YourConfig): Adapter {
  return new Adapter(
    "your_client_name",
    ({ messages, schema }) => {
      return {
        resolve() {
          // Request inference from model.
          // Return the inferred value.
        },
        stream() {
          // Request inference from model with streaming enabled.
          // Return a ReadableStream<string>
        },
      }
    },
  )
}