HolaOS — Proactive AI Agent That Learns Your Workflow

Introduction

HolaOS is an agent operating system that sits alongside your development workflow, continuously learning your patterns, project context, and preferences. Unlike chat-based assistants that start fresh each session, HolaOS maintains persistent memory and proactively surfaces relevant information or suggests actions based on what it has learned about how you work.

What HolaOS Does

• Observes your development activity to build persistent context
• Maintains long-term memory across sessions and projects
• Proactively suggests actions based on learned workflow patterns
• Answers questions about your codebase, recent changes, and decisions
• Integrates with editors, terminals, and browser via MCP protocol

Architecture Overview

HolaOS runs as an Electron desktop app with a background TypeScript runtime. The runtime monitors configured workspaces for file changes, git activity, and terminal commands. Events are processed into a structured memory store using embedding-based indexing. A proactive engine periodically evaluates the current context against learned patterns and surfaces suggestions. Communication with LLM providers happens through a local proxy that manages API keys and context windows.

Self-Hosting & Configuration

• Available as an Electron app for macOS, Windows, and Linux
• Configure watched workspaces and data retention in the settings panel
• Supports any OpenAI-compatible LLM provider including local Ollama endpoints
• Memory data stored locally in ~/.holaos/ with optional encryption
• Privacy controls let you exclude specific directories or file types from observation

Key Features

• Persistent cross-session memory that never forgets your project context
• Proactive suggestions surfaced via system notifications
• Multi-workspace support for switching between projects
• Privacy-first design with all data stored locally
• MCP-compatible for integration with coding agents and IDE extensions

Comparison with Similar Tools

• GitHub Copilot — reactive code completion; HolaOS is proactive and context-persistent
• Cursor — IDE with AI features; HolaOS is IDE-agnostic and focuses on workflow memory
• Khoj — personal AI second brain; HolaOS emphasizes proactive suggestions over search
• Mem0 — memory layer for LLMs; HolaOS is a full agent runtime, not just a memory API

FAQ

Q: What data does HolaOS observe? A: File changes, git commits, and terminal commands in configured workspaces. Browser activity and communication apps are never monitored.

Q: How much storage does the memory use? A: Typical projects generate 10-50MB of memory data per month. Retention policies can be configured to auto-prune older entries.

Q: Can I use it without cloud LLMs? A: Yes, HolaOS works fully offline with Ollama. The proactive engine adapts its suggestion quality to the capability of the local model.

Q: Is the observed data sent anywhere? A: No. All observation data stays in your local ~/.holaos/ directory. LLM queries include only the relevant context snippet, not your full history.

Sources

• https://github.com/holaboss-ai/holaOS