getomnico/omni

What it solves

Omni provides a centralized, self-hosted AI agent for companies to manage work across fragmented software tools. It eliminates the need to manually search through multiple platforms like Slack, Jira, and Google Drive by allowing employees to investigate issues, analyze data, and execute tasks from a single conversation, all while respecting the original permissions of the source systems.

How it works

Omni connects to various workplace tools via connectors and indexes company data into a Postgres database using ParadeDB for full-text search and pgvector for semantic search. It uses a Python-based agent runtime to orchestrate models and execute tasks. For data analysis and file inspection, it employs a sandboxed environment to run Python and bash code securely. The system is designed to be self-hosted via Docker Compose or Terraform, supporting a wide range of LLM providers including OpenAI, Anthropic, and local options like Ollama.

Who it’s for

Companies that need an AI assistant capable of accessing internal company data across multiple third-party tools but require the security and control of a self-hosted deployment.

Highlights

• Permission-Aware Context: Ensures indexed information is scoped to the user's existing permissions in the source systems.
• Hybrid Search: Combines BM25 and semantic search within a single Postgres instance.
• Sandboxed Execution: Runs code in isolated containers with Landlock filesystem restrictions for secure data analysis.
• Broad Integration Support: Out-of-the-box connectors for Google Workspace, Microsoft 365, Slack, Jira, Confluence, and more.
• Model Agnostic: Compatible with major cloud LLMs and OpenAI-compatible endpoints.