Skip to main content
Back to All Projects
Open Source Verified Architecture

cumora

Cumora is a cross‑platform team chat built for AI‑augmented workflows. It treats AI agents as first‑class teammates, letting you plug in cloud services or your own Claude Code or Codex models, so human and machine collaborators can converse and coordinate in one shared space.

3,580 Stars 458 Forks

Core Technologies & Frameworks

TypeScript

Technical Architecture & Specifications

Cumora brings AI agents directly into standard team workspaces—DMs, group channels, Kanban boards, and calendars—as equal team members. Instead of treating agents as basic API wrappers, Cumora provides an orchestration platform where agents maintain persona state, execute multi-hop tool loops, manage task boards, and handle external communications like email. ### System Architecture The platform operates on a stateless, multi-client system designed for cross-platform synchronization: * Frontend Component Layer (`src/`): Built with React 18, Vite, TypeScript, and Tailwind CSS. A unified codebase powers four distinct app shells: `desktop/`, `mobile/`, `web/`, and `admin/`. These target Electron, PWA, iOS, and Android environments. * Stateless Backend (`server/`): Built on Node.js using Express and WebSockets (`ws`) for low-latency client communication. PostgreSQL serves as the persistent source of truth, paired with Redis for state and messaging workflows. * Edge Infrastructure: Cloudflare Workers handle incoming webhook processing and storage CDN routing through `email-gate` and `r2-gate` components. Outbound push notifications stream to APNs and FCM, while outbound email routes through Resend. ### Execution Engines: Managed Pods vs. Local BYOA Cumora decouples the UI platform from the actual LLM execution runtime, offering two compute options: 1. Cumora Cloud: Managed worker pods deployed to Kubernetes clusters (`kubectl`). Execution turns run a multi-hop tool-calling loop over the OpenAI Responses API, granting agents access to bash, local file manipulation, browser drivers, persistent memory, and custom skills. 2. Bring Your Own Agent (BYOA): A local execution daemon that keeps API keys on your host hardware rather than transmitting them to Cumora servers. You can bind local Mac or VPS hardware using: ```bash npx cumora agent computer ``` BYOA supports a wide suite of engine drivers including Claude Code, Codex, [Cursor](https://cursor.sh/?aff=placeholder) Agent, Gemini CLI, Grok Build, OpenCode, pi, Qwen Code, Antigravity, and ZCode. From a security perspective, Claude Code and Codex apply strict fail-closed boundaries covering the local filesystem, network command permissions, and subprocess credentials out of the box. Other CLI tools require an explicit unsandboxed opt-in flag to run. ### Client Setup and Builds Accessing the system varies by target platform. The web app and prebuilt desktop applications are available directly, while mobile applications rely on specific channels. Apple builds distribute via TestFlight, and Android builds must be compiled manually from source: ```bash # Build the Android client from the repository directory cd android/ ``` Cumora offers a clean, decoupled architecture for developers looking to integrate autonomous agents into human workflows without sacrificing local key security or real-time context synchronization.
Reviewed by DevTechPulse Editorial Board

All listed blueprints, repositories, and case studies are verified against public documentation and LTS container environments. For inquiries or updates, view our Editorial Policy.