@c3-oss/codexcw

Run Codex, Claude Code, or Grok Build non-interactively from Node.js. The Codex agent wraps codex exec --json; Claude wraps claude -p --output-format stream-json; Grok uses --no-auto-update and streaming-messages-json. All expose results, live event streams, callbacks, typed usage, and typed errors through the same API. A native addon backed by a Rust core (napi-rs) provides prebuilt binaries for macOS, Linux (gnu + musl), and Windows.

Install

npm install @c3-oss/codexcw

The selected agent executable must be on PATH and authenticated. Codex must support codex exec --json; Claude must support --output-format stream-json; Grok must support streaming-messages-json.

The Claude agent is selected with new Runner({ agent: 'claude' }); its events are normalized into the same event model, with model selection via the haiku/sonnet/opus aliases (ClaudeModel).

Select Grok with new Runner({ agent: 'grok' }). New runs use read-only sandboxing and dontAsk permissions; Grok sessions are always persisted.

Usage

The API is async-only: every call returns a Promise, and event streams are AsyncIterables consumed with for await.

import { Runner } from '@c3-oss/codexcw'

const runner = new Runner()

const result = await runner.run({ prompt: 'say hi' })
console.log(result.finalMessage)
console.log(result.usage.totalTokens)
console.log(result.usage.totalCostUsd)

import {
  getAccountUsage,
  getClaudeAccountUsage,
} from '@c3-oss/codexcw'

const usage = await getAccountUsage({
  env: { CODEX_HOME: '/tmp/codex-home' },
})
console.log(usage.rateLimits.primary?.usedPercent)
console.log(usage.tokenUsage?.summary.lifetimeTokens)

const claudeUsage = await getClaudeAccountUsage()
console.log(claudeUsage.windows)

Full recipes — streaming, resume, sandbox/approval, batches, structured output, account usage, and error handling — are in docs/examples/typescript.md.

License

CC0-1.0