Omnara is the open-source API for creating and interacting with production-ready AI agents. It provides durable execution, state management, and the underlying infrastructure, so developers don’t have to build their own agent control plane. Choose your models, tools, and execution environments, with the flexibility to use managed hosting or self-host in your own environment.
Key Features
Unified agent API: Create, configure, and interact with AI agents through a standard API. Integrate agents into your applications without building a separate agent management layer.
Durable execution and state: Preserve agent sessions and conversation history so work can resume after interruptions without depending on a single running process.
Bring your own models: Use built-in support for any model from model providers like OpenAI, OpenRouter, Anthropic, and Amazon Bedrock with the option to bring your own API key. Or connect any compatible LLM endpoints. Choose the models that fit your needs without being restricted to one provider.
Bring your own machines and sandboxes: Connect local computers, cloud VMs, or any environment capable of running Omnara’s lightweight daemon. Access supported sandbox providers through Omnara or bring your own provider API keys.
Custom tools, skills, and MCP integrations: Equip agents with your own instructions and capabilities, and connect the services they need to complete tasks.
Managed hosting or self-hosting: Use Omnara’s hosted service or deploy the open-source control plane in your own cloud or on-premises environment.
Serverless agent architecture: Keep agent state independent of execution compute, without maintaining a dedicated, always-on server for every agent.
Benefits
Lower compute and model costs: Use serverless execution to reduce idle compute costs, and choose cost-effective models for each task instead of defaulting to the most expensive frontier models.
Keep your agent data under your control: Self-host the control plane in your own cloud or on premises to control where agent conversations, logs, and state are stored, who can access them, and how long they are retained.
Spend more time building useful agents: Focus on agent capabilities and your application instead of building the APIs, session storage, recovery logic, and supporting infrastructure needed to run them in production.
Turn agent history into a long-term asset: Use retained logs and interaction history to evaluate performance, improve agent behavior, and prepare datasets for training or fine-tuning your own models.
Keep your stack flexible: Choose models, tools, and execution environments independently. Adapt your setup as requirements change rather than committing every layer to one provider.
Put existing resources to work: Use your own model API keys, local computers, cloud machines, and supported sandbox accounts rather than replacing infrastructure you already have.
