schmitech/orbit

What it solves

ORBIT is a self-hosted AI backend designed to bridge the gap between applications and the diverse set of models, data sources, and tools required for private, governed AI agents. It eliminates the need to rewrite architecture when moving from a local prototype to a production deployment by providing a unified, OpenAI-compatible API for managing private data and tool-using agents.

How it works

ORBIT acts as an intermediary layer that routes requests to various inference providers (such as Ollama, vLLM, OpenAI, or Anthropic) while connecting to data sources via YAML-configured adapters. These adapters allow the system to integrate files, SQL/NoSQL databases, vector stores, Elasticsearch, and REST/GraphQL APIs. It also supports Model Context Protocol (MCP) tools for agentic capabilities. The entire system is managed through a built-in admin panel and YAML configuration files for global settings, inference providers, and adapter behavior.

Who it’s for

It is intended for developers and organizations that need to deploy AI agents with strict control over their data and infrastructure, including those operating in on-premises, private cloud, or air-gapped environments.

Highlights

• Unified API: Provides a single OpenAI-compatible API contract regardless of the underlying model provider.
• Extensive Connectivity: Connects to a wide array of data sources including SQL, NoSQL, vector stores, and APIs via adapters.
• Enterprise Governance: Includes built-in RBAC, OIDC/SSO, audit logs, file encryption, and per-key quotas.
• Production Resilience: Features provider fallbacks, circuit breakers, health checks, and hot adapter reloads.
• Admin Control Plane: An integrated UI for managing API keys, prompts, and adapters without modifying server code.