Install

npm install theazo

Quick Start

import { Theazo } from 'theazo'

const theazo = new Theazo({ apiKey: process.env.THEAZO_API_KEY })

// Session per user — isolated compute, billing, cost limits
const session = await theazo.sessions.forUser('user_123')

// One-liner: sandbox + Claude + tools + cost tracking
const result = await session.run('researcher', 'analyze competitor pricing')

console.log(result.output)   // "Based on analysis of 12 competitors..."
console.log(result.cost)     // { amount: 3, currency: 'usd' }

AI Infrastructure

The operating layer every AI agent company needs — included out of the box.

| | | |---|---| | Sessions | Per-user isolation — User A can't see User B's data, agents, or costs | | Agents | Sandboxed Python/Node/Go compute. Managed (E2B) or bring your own | | Billing | Per-user cost tracking, 4-level budget caps (platform/user/session/run), threshold alerts | | Observability | Real-time logs, traces, cost breakdown per agent and per user | | Cost Controls | Auto-pause when limits hit. No surprise bills. | | Streaming | Token-by-token SSE for real-time agent output | | Model Gateway | 500+ models. Managed or BYOI (direct key, OpenRouter, LiteLLM, Azure, vLLM, Ollama) | | Secrets Vault | AES-256-GCM encrypted credentials. Injected at runtime. Never logged. |

AI Primitives

Agent orchestration building blocks. Your product, Theazo's backend.

| | | |---|---| | Workflows | Multi-step pipelines — 9 step types, parallel branches, conditional logic, dynamic DAGs | | Fleets | Batch N tasks in parallel with concurrency control and cost caps | | Chat | Multi-turn conversations with streaming, context strategies, threads | | Knowledge / RAG | Upload docs → agents query them automatically via pgvector | | MCP Tools | Connect external tool servers — agents discover and use tools automatically | | Channels | Chat widget, Slack, email — agents meet users where they are | | Scheduling | Cron schedules + webhook triggers for recurring agent tasks | | Approvals | Human-in-the-loop — pause for approval before sensitive actions | | Agent Store | Save, version, rollback, duplicate agent definitions | | Guardrails | PII detection, prompt injection protection, content filtering |

Three Modes

Fully Managed — Theazo handles compute, models, and orchestration:

const result = await session.run('researcher', 'analyze pricing')
// compute + model + tools + billing — all handled

Infra-Only — Run your own agent code (LangGraph, CrewAI) in a Theazo sandbox:

const agent = await session.agents.create({ compute: 'python' })
const result = await agent.exec('python', myLangGraphCode)
// You run the agent loop. Theazo handles compute, isolation, billing.

BYOI — Bring your own compute and/or models:

// Configure via dashboard or REST API — then SDK works the same
const result = await session.run('researcher', 'analyze market')
console.log(result.cost) // orchestration only — compute + model = $0

Supports: E2B, Fly.io, custom K8s (compute) + Anthropic, OpenAI, OpenRouter, LiteLLM, Azure, vLLM, Ollama (models).

Workflows

import { Theazo, workflow } from 'theazo'

const pipeline = workflow('research-report')
  .step('research', { agent: 'researcher', input: { topic: '$.trigger.topic' } })
  .step('analyze', { agent: 'analyst', input: { data: '$.research.output' } })
  .step('write', { agent: 'writer', input: { analysis: '$.analyze.output' } })
  .build()

const wf = await theazo.workflows.create(pipeline)

for await (const event of theazo.workflows.streamRun(wf.id, { input: { topic: 'AI agents' } })) {
  if (event.event === 'step.completed') console.log(`✓ ${event.stepId} — $${event.cost / 100}`)
}

Step types: agent · parallel · condition · transform · map · delay · approval · webhook · planner (dynamic DAG — agent decides what steps to add at runtime)

Fleets

const fleet = await session.fleets.dispatch({
  agent: 'researcher',
  inputs: companies.map(c => ({ company: c })),
  concurrency: 5,
})

for await (const item of fleet.stream()) {
  console.log(`Done: ${item.input.company} — $${item.cost / 100}`)
}

Chat + Streaming

// Multi-turn conversation
const conv = await session.chat.create({ agent: 'assistant' })
const reply = await session.chat.send(conv.id, { content: 'What is Theazo?' })

// Stream tokens
for await (const event of session.chat.stream(conv.id, { content: 'Write a haiku' })) {
  if (event.type === 'text') process.stdout.write(event.text)
}

Knowledge / RAG

await session.knowledge.upload({
  collection: 'docs',
  source: 'text',
  content: 'Theazo pricing: Free $0, Pro $99/mo...',
  filename: 'pricing.md',
})

// Agent uses knowledge automatically
const result = await session.run('assistant', 'What does Pro cost?', {
  knowledge: 'docs',
})

MCP Tools

const conn = await theazo.mcp.connect({
  name: 'github',
  transport: 'sse',
  url: 'https://mcp-github.example.com/sse',
})
// Agent discovers: github_create_issue, github_search_repos, ...

const result = await session.run('developer', 'Find open bugs in our repo', {
  mcp: [conn.id],
})

Examples

13 runnable examples at github.com/theazo/theazo-node/examples:

basic-agent · streaming · chat · workflows · fleet-dispatch · agent-definitions · mcp-tools · approvals · byoi · knowledge · scheduling · channels · infra-only

Documentation

theazo.com/docs — Getting started, API reference, BYOI guide, workflows, billing.

Requirements

Node.js 18+ · TypeScript 5+ (recommended)

License

MIT