Building AI-Powered CLI Tools with Node.js: Zero-Config Streaming Pipelines

[chawkian]

CLI tools are where Node.js shines for AI workflows. Fast startup, universal runtime, and process.stdout is basically made for streaming.

The Pattern

Instead of building a chatbot UI, pipe LLM output directly through Unix tools:

npx aiwave chat "Summarize today's HN front page" | tee summary.md | glow -

Minimal Streaming CLI

import { AIWaveClient } from '@aiwave/sdk-node';

const client = new AIWaveClient();

const stream = await client.chat.stream({
  model: 'glm-5',
  messages: [{ role: 'user', content: process.argv[2] }],
});

for await (const chunk of stream) {
  process.stdout.write(chunk.delta ?? '');
}

That's it. 12 lines, zero deps beyond the SDK, and you've got a streaming AI command.

Pro Tips

• Use stdin for context. Pipe in a file, a curl output, a database dump — the LLM sees it all.
• process.stderr for metadata. Token counts, latency, model info — route to stderr so stdout stays pure output.
• Exit codes matter. If the API fails mid-stream, exit 1 so shell pipelines (set -e) catch it.

cat error.log | npx aiwave chat "What's causing this crash?" 2>meta.txt

When to Go Heavier

For multi-step agents (tool calling, RAG pipelines), check the AIWave Node SDK guide — it covers the agent loop pattern with structured tool outputs.

The beauty of CLI-first design is that composability is free. Your AI tool works with grep, jq, pbcopy, and every other Unix tool invented in the last 50 years.

---

What AI CLI tools have you built? Has anyone shipped a git integration that auto-generates commit messages from diffs? Share your pipelines.

---

Resources

• AIWave pricing
• AIWave developer blog
• AIWave Node.js SDK