@llm4ts/runner

Node runner for llm4ts: the runNode embedded entry point, connector presets, terminal rendering, the llm4ts CLI, and the MCP stdio server.

Install

npm install @llm4ts/runner @llm4ts/flow @llm4ts/core effect

Requires Node.js >= 22. effect is a peer dependency pinned to the Effect 4 beta line.

Run a flow

runNode provides every Node boundary (HTTP client, process executor, temporary files, file store, connector registry) with sensible defaults — you never assemble Effect layers yourself:

import { runNode } from "@llm4ts/runner/FlowRunner"
import { ApiConnectorConfig } from "@llm4ts/core/ConnectorConfig"
import { ConnectorIds } from "@llm4ts/core/Models"
import { collect } from "@llm4ts/core/Streaming"

const program = runNode(
  {
    workDir: process.cwd(),
    workspace: process.cwd(),
    userPrompt: "Draft a small implementation plan",
    coder: ApiConnectorConfig.make({ connectorId: ConnectorIds.Mock })
  },
  (context) => collect(context.coder.executeStream(context.userPrompt))
)

The mock connector needs no credentials. Swap in a real provider via ApiConnectorConfig (API keys are read from standard environment variables such as ANTHROPIC_API_KEY) or a CLI coding agent via coderFromEnv (LLM4TS_CODER=claude|codex|gemini|...).

CLI

The package ships a llm4ts binary that streams one prompt to your configured coder and records a trace under .llm4ts/:

npx llm4ts "explain this repository" --repo .

Documentation

• Repository and full documentation
• Configuration guide
• Runnable examples

License

MIT