OpenMolt

About

OpenMolt lets you spin up autonomous AI agents in Node.js that go beyond conversation — they read your Gmail, triage GitHub issues, post to Slack, and manage Stripe payments, all through a single TypeScript config. The first time you define an agent in 15 lines of code and watch it autonomously pull metrics from three APIs and summarize them in Slack, you realize how much boilerplate you were writing before. The framework ships with 30+ built-in integrations covering Gmail, Slack, GitHub, Notion, Stripe, Discord, S3, Google Workspace (Calendar, Drive, Sheets), Shopify, Airtable, Twilio, Instagram, X, YouTube, Dropbox, and browser automation. Each integration uses declarative HTTP tool definitions with Liquid template rendering — you describe what the tool does as data, not code. No writing fetch calls or parsing responses. Security is where OpenMolt actually stands out from similar frameworks. It runs a zero-trust model: API credentials stay server-side and never get passed to the LLM. The model only sees tool results, not raw tokens or secrets. Scopes gate which tools each agent can access, so your email-reading agent cannot accidentally trigger a Stripe refund. You pick your LLM backend with a simple provider:model string — OpenAI GPT-4o, Anthropic Claude, or Google Gemini — and switch between them without rewriting agent logic. Structured output via Zod schemas means your agent returns typed, validated JSON instead of hoping the LLM formatted things correctly. For recurring workflows, built-in scheduling supports interval-based and daily cron-style execution with timezone awareness. The memory system provides both short-term (conversation context) and long-term (persistent) storage with custom callbacks for your own database. The honest downside: OpenMolt is early-stage with 26 GitHub stars and a small community. Documentation exists but is thin compared to LangChain or CrewAI. If you need battle-tested production reliability with enterprise support, this is not there yet. But if you are a developer who wants a clean, opinionated TypeScript framework for building real AI automations without the abstraction bloat of larger frameworks, OpenMolt is worth a serious look. Created by Youssef Bouane, MIT licensed, and actively maintained.

Key Features

• 30+ built-in integrations (Gmail, Slack, GitHub, Notion, Stripe, Discord, S3, Shopify, and more)
• Multi-LLM support via unified provider:model syntax (OpenAI, Anthropic Claude, Google Gemini)
• Zero-trust security — API credentials never exposed to the LLM, scope-gated tool access
• Structured output with Zod schema validation for typed, reliable responses
• Declarative HTTP tool definitions using Liquid templates — no boilerplate fetch code
• Interval and cron-style scheduling with timezone support
• Long-term and short-term memory systems with custom persistence callbacks
• CLI mode — run agents from JSON config files via npx openmolt agent.json
• Observable event system for tool:call, tool:response, finish, and planUpdate hooks
• System prompts loadable from Markdown files via instructionsPath

Use Cases

• 1Daily reporting agents that pull metrics from multiple APIs and post summaries to Slack
• 2Email management bots that read, classify, and draft replies in Gmail
• 3GitHub issue triage — auto-label, assign, and respond to new issues
• 4Content pipelines that generate text, create images, and store assets in S3
• 5E-commerce automation with Shopify order monitoring and Stripe payment handling
• 6Customer research agents that aggregate data from Airtable, Notion, and Google Sheets
• 7Scheduled competitive analysis bots that run daily and report findings to Discord