veryfront/llm

Import

ts
import { fileFromBytes, generate, planFileParts } from "veryfront/llm";

Examples

Text

ts
import { generate } from "veryfront/llm";

const { text } = await generate({ input: "Name three colours." });

Structured output

ts
import { generate } from "veryfront/llm";
import { defineSchema, lazySchema } from "veryfront/schemas";

const { object } = await generate({
  input: "The checkout button does nothing on mobile Safari.",
  system: "Classify the support ticket.",
  outputSchema: lazySchema(defineSchema((v) =>
    v.object({
      category: v.enum(["bug", "billing", "feature"]),
      reasoning: v.string(),
      confidence: v.number().min(0).max(100),
    })
  )),
});

object.category; // "bug" | "billing" | "feature"
object.confidence; // number, 0-100

Choosing a model

ts
import { generate } from "veryfront/llm";

const { text } = await generate({
  input: "Summarise this in one line.",
  system: "You are terse.",
  model: "anthropic/claude-haiku-4-5-20251001",
});

Exports

Functions

NameDescriptionSource
fileFromBytesBuild a serializable message file part without uploading the file.source
generateRun a single model call without registering a reusable agent.source
planFilePartsSelect which local files fit a message, without throwing for file data.source

Types

NameDescriptionSource
FileBytesInputLocal file input. Read bytes with your runtime's filesystem or File API.source
FilePartOmissionReasonWhy planFileParts did not send a file.source
FilePartsPlanFile parts to send and every file that was left out.source
GenerateInputRequest accepted by generate.source
OmittedFilePartA file that planFileParts did not send. File bytes are not included.source
PlanFilePartsInputLocal files and optional raw-byte and count budgets for one message.source