How Omni works

One system your team can deploy and operate

Omni brings the runtime, integrations, company context, model access, and permissions into one open-source system designed to run together.

Get early access View on GitHub

The capabilities behind a company-wide agent

Teams can build these capabilities, buy them, or combine several systems. However they are provided, they have to work together.

Agent runtime

Orchestration, task state, tools, sandboxing

Plans and executes work, invokes tools, maintains task state, and runs code safely.

Tool integration

MCP, direct APIs, integration services

Makes company systems available to the agent while handling authentication, schemas, errors, and source API limits.

Model access

Provider SDKs, gateways, managed endpoints

Connects the agent to the models your company uses and handles credentials and provider differences.

Search and retrieval

Search engines, vector stores, managed retrieval

Indexes company information and retrieves the context needed for each task.

Content sync and actions

Connectors, MCP servers, custom integrations

Keeps company data current and carries out actions while handling OAuth, pagination, rate limits, and schema changes.

Identity and authorization

Source permissions, SSO, policy checks

Ensures each employee can only retrieve data and perform actions they are allowed to access.

How Omni handles each layer

Omni includes the runtime, tool access, model support, connectors, searchable company context, permissions, and governance in one open-source system you can run in your own environment.

Capability
Agent runtime Common:Frameworks or custom orchestration code Built-in runtime for tools, task state, MCP, and sandboxed code execution
Tool integration Common:MCP servers, direct APIs, or integration services Native MCP support and connector actions through the same agent interface
Model access Common:Provider SDKs or a model gateway Support for Anthropic, OpenAI, Gemini, Azure, Bedrock, and OpenAI-compatible endpoints
Search and retrieval Common:A search engine, vector store, or managed retrieval service BM25 and semantic retrieval with ParadeDB and pgvector in Postgres
Content sync and actions Common:ETL tools, connector services, or custom integrations Maintained workplace connectors plus Python and TypeScript connector SDKs
Identity and authorization Common:Source APIs, ACL synchronization, or a policy engine Source permissions carried into indexed context and connector actions
Operations and governance Common:Deployment tools, audit pipelines, identity services, and cost reporting Docker Compose and Terraform deployment, audit logs, SSO/SCIM, and cost tracking

Start with Docker Compose, or use the Terraform modules for production deployments on AWS or GCP.

Run Omni in your own environment

Start with Docker Compose, explore the production deployment guides, or join the early access list for Omni Cloud.

Star on GitHub Read the docs