mediar-ai/fazm

Fazm Desktop for macOS

What it solves

Fazm tackles repetitive computer work by letting you control your browser, write code, handle documents, and operate Google Apps through voice commands. It also learns your workflows so tasks become more automated over time.

How it works

Fazm is a macOS app built with Swift and SwiftUI in the Desktop folder, plus an ACP bridge written in TypeScript for Claude integration. Running ./run.sh builds and launches the app. On a fresh clone, it fetches a small non-secret bootstrap config from the hosted Fazm backend (backend URL, public API keys, OAuth client ID, GCP identifiers) and bakes that into the app bundle. Billable model keys are fetched at runtime from the backend, gated by authentication. If you prefer your own backend, you can place a .env.app (or .env.app.dev) file at the repo root with your own values, and run.sh will use it instead.

Who it's for

Fazm is aimed at anyone who wants to offload computer tasks to an AI agent using natural voice commands — from developers to professionals handling documents, spreadsheets, and customer relationships. Because it's free to start and fully open source, it appeals to both individual users and teams looking for a local, customizable automation tool.

Highlights

  • Voice control for browser, code, documents, and Google Apps
  • Learns your workflow to repeat tasks automatically
  • Fully open source and free to start
  • Runs locally on macOS (requires macOS 14.0+)
  • Includes demos for Smart Connections, CRM Management, and visual task understanding
  • Visual context awareness for complex workflows
  • Can connect with the right people across platforms automatically
  • Keeps CRM data up to date without manual entry

Related

  • Project
    egoist/lorca

    Lorca is a self-hosted, end-to-end encrypted bot orchestration platform that runs bots on your own machines and lets them collaborate in group chats. It includes desktop apps, a CLI, and mobile apps, supporting multiple AI providers and offering a private alternative to cloud bot services.

  • Project
    Zero-Mem/Zero-mem

    Zero-Mem is a memory framework for LLM agents that performs all memory operations (construction, organization, retrieval, calibration) without any LLM calls or tokens, using only an encoder and NER for computation. It builds an entity–context graph and temporal hierarchy over original interaction traces, retrieves evidence via query-conditioned routing, and uses a reader LLM only for final answer generation. Reported results show GPT-4o-mini achieving 59.15 F1 on LoCoMo and 72.07 F1 on HotpotQA 56K, with 57.6% lower memory-operation time than the LightMem baseline.

  • Project
    yjh051108/dsh-router-standard

    A research‑level router preset for DeepSeek Harness that automatically classifies the first user request and injects the appropriate persona and tool set (spec/react/weak) for the DeepSeek V4 model. It runs as a zero‑dependency Cord‑is plugin, integrates with the dsh‑super‑injector for hot‑reloading, and provides CLI diagnostics. Currently in development (v1.19.1, Aug 2026).

  • Project
    ferdousbhai/investor-agent

    An MCP server that provides AI assistants with tools to fetch real-time financial data, stock fundamentals, and technical indicators for investment research.