KIRA  by krafton-ai

On-premise AI agent for desktop productivity

Created 1 month ago
513 stars

Top 61.0% on SourcePulse

GitHubView on GitHub
Project Summary

KIRA (Krafton Intelligence Rookie Agent) is an open-source desktop application designed as a 24/7 AI virtual coworker. It targets individuals and teams seeking an AI assistant that operates entirely locally, ensuring data privacy and eliminating the need for server setup or cloud infrastructure. KIRA integrates with communication and productivity tools like Slack, Outlook, Confluence, and Jira, acting as a proactive assistant to streamline workflows and manage tasks.

How It Works

KIRA utilizes a multi-agent architecture where specialized agents handle distinct roles, enabling cost optimization by employing lighter models for simpler tasks. A sophisticated message processing pipeline, including debouncing and bot call detection, routes messages to appropriate agents. A three-tier queue system (Channel, Orchestrator, Memory) manages concurrent operations efficiently: Channel queues provide fast, independent responses per channel; the Orchestrator queue handles heavy tasks with limited workers to prevent resource overload; and the Memory queue ensures sequential, conflict-free writes to local storage. This design prioritizes responsiveness, resource management, and data integrity.

Quick Start & Requirements

  • Prerequisites: macOS 10.15 or later, Slack workspace with admin access, Claude API key (or Google Cloud Vertex AI).
  • Installation: Download the latest release from https://kira.krafton-ai.com or build from source by cloning the repository (https://github.com/krafton-ai/kira.git), installing Python dependencies (uv sync), building the Electron app (npm install, npm run build), and installing the generated .dmg file.
  • Configuration: Requires creating a Slack App, launching KIRA, entering Slack tokens and bot information in the Settings, and starting the application.
  • Documentation: Comprehensive guides are available via the vitepress-app/ directory or linked resources.

Highlighted Details

  • Core Functionality: Natural language chat in Slack (DMs, channels, threads), task execution (web search, document writing, email management, Jira/GitLab integration), local memory system for context retention, and task scheduling.
  • Proactive Features (Beta): Email monitoring (Outlook, 5-minute intervals), Confluence document update notifications (1-hour intervals), Jira issue tracking (30-minute intervals), and proactive task suggestions based on memory analysis.
  • Integrations: Supports Perplexity (web search), DeepL (translation), Outlook, Confluence, Jira, GitLab, X (Twitter), Clova Speech (transcription), and Playwright (browser automation).
  • Privacy: All data and memory are stored locally on the user's computer, with no data collection by KRAFTON.

Maintenance & Community

Contributions are welcomed via pull requests on the GitHub repository. Specific community channels or details on core maintainers/sponsorships are not detailed in the README.

Licensing & Compatibility

  • License: Apache License 2.0.
  • Compatibility: Primarily designed for macOS. The local-first approach is suitable for environments prioritizing data privacy. Users are responsible for adhering to third-party API terms (e.g., Anthropic Claude API, Slack). KRAFTON's trademarks require separate written permission for use.

Limitations & Caveats

The Proactive Monitors and Proactive Suggestions features are currently in Beta. The application requires macOS 10.15 or later. Users bear full responsibility for managing their API keys, associated costs, and security for all integrated third-party services. KRAFTON explicitly disclaims liability for any data processed or actions taken by these external services.

Health Check
Last Commit

2 days ago

Responsiveness

Inactive

Pull Requests (30d)
29
Issues (30d)
4
Star History
511 stars in the last 30 days

Explore Similar Projects

Feedback? Help us improve.