Let your code create and manage AI Agents (OpenSource)

OpenMolt provides a complete toolkit for developers building production-grade autonomous AI agents. Unlike no-code platforms, OpenMolt integrates directly into your existing Node.js and TypeScript codebase, giving you full programmatic control over agent behavior, reasoning, and tool execution. The framework emphasizes security-first design with scope-based permissions, ensuring API credentials never leave your server while allowing agents to interact with 30+ popular services.

The platform serves developers and engineering teams who need reliable, observable AI automation for business workflows. Common use cases include automated reporting pipelines, content generation workflows, email management, GitHub automation, e-commerce operations, and customer research. OpenMolt is particularly valuable for teams that require recurring scheduled tasks, multi-step reasoning with memory persistence, and integration with existing business tools like Slack, Notion, Stripe, and Google Workspace.

• Secure by Design: Agents access only the scopes you explicitly grant, with API credentials stored server-side and never exposed to LLMs — the LLM only sees tool names, never raw credentials.

• Multi-Provider LLM Support: Use OpenAI GPT-4o, Anthropic Claude, or Google Gemini with a unified model string format, allowing you to switch providers without changing your code.

• 30+ Built-in Integrations: Connect to Gmail, Slack, GitHub, Notion, Stripe, Discord, Amazon S3, and more — all ready to use out of the box with zero configuration.

• Structured Output with Zod: Pass a Zod schema and receive validated, typed objects in return, eliminating manual parsing and validation of LLM responses.

• Flexible Scheduling: Schedule agents with interval-based or cron-style daily schedules with timezone support, perfect for recurring reports and automated workflows.

• Event-Driven Architecture: Hook into every step of the reasoning loop to observe tool calls, plan updates, LLM outputs, and results in real-time.

• Persistent Memory System: Long-term and short-term memory stores with onUpdate callbacks allow agents to learn and remember context across multiple runs.

• Declarative Tool Definition: Define integrations as data using endpoints, auth templates, and schemas — no boilerplate HTTP code required.

• Install the package: Run npm install openmolt in your Node.js project to add the framework to your dependencies.

• Configure LLM providers: Initialize OpenMolt with your API keys for OpenAI, Anthropic, or Google Gemini in the llmProviders configuration object.

• Create an agent: Use om.createAgent() to define your agent with a name, model, instructions, and any additional configuration like output schemas or scope restrictions.

• Run your agent: Call await agent.run() with your input prompt to execute the agent and receive the result, which will include any tool calls and structured output.

• Add integrations: Enable specific integrations by granting scopes to your agent, allowing it to access services like Gmail, Slack, or Notion through declarative tools.

• Schedule recurring tasks: Configure interval or cron-based schedules to run your agent automatically for reports, monitoring, or other recurring workflows.

• Implement memory persistence: Provide an onUpdate callback to save agent memory to your preferred database or file system across sessions.

• Zero-Trust Security Model: OpenMolt's scope-based permission system ensures API credentials are rendered into HTTP requests via Liquid templates server-side, meaning the LLM never receives raw API keys or tokens — only tool call results.

• Fully Configurable Agents: Every aspect of an agent is customizable, including instructions, model provider, model configuration parameters, output schemas, maximum steps, and per-integration scope restrictions.

• Production-Ready Observability: The event-driven architecture provides hooks into every step of the reasoning loop, enabling real-time monitoring, debugging, and logging of agent behavior.

• Seamless Provider Switching: The unified model string format allows you to switch between OpenAI, Anthropic, and Google models without refactoring your codebase.

• Memory Persistence Across Sessions: Agents maintain both long-term and short-term memory with callbacks for custom persistence, enabling context-aware interactions over time.

• Documentation: Comprehensive documentation is available at openmolt.dev/docs covering installation, configuration, API reference, and integration guides.

• Playground: An interactive web-based playground at openmolt.dev/playground allows you to configure and test agents without writing code.

• GitHub Repository: Report issues, request features, and contribute to development at the GitHub repository.

• Community: Follow updates and engage with the community on Twitter / X for announcements and support.

• NPM Package: Install via npm install openmolt — requires Node.js environment with TypeScript support recommended.

• Web Playground: Accessible directly in browser at openmolt.dev/playground — no download required for testing and experimentation.