# Iron-Jarvis

> Local-first desktop app where AI agents plan, build and review code you approve

- Repository: https://github.com/RealDealCPA-VR/Iron-Jarvis
- Category: AI agents
- Stars: 63 · Forks: 17 · Language: Python · License: not stated
- Experience: Some experience
- Page: https://ir.adyog.com/repos/realdealcpa-vr/iron-jarvis/

Iron Jarvis tries to be an operating system for AI agents rather than a chatbot. You type into one chat surface. Quick questions get a fast answer, and requests that need real tool work are handed to a full agent, with a visible reason.

A supervisor splits a goal across subagents such as Planner, Builder, Reviewer and Researcher. Each works in a disposable git worktree. Code lands on a branch and is never merged automatically, so you read the diff and approve it. Every tool call passes a permission engine that denies by default.

It can use cloud models through an API key or an existing claude/codex CLI login, or local servers such as Ollama, vLLM and LM Studio. A deterministic mock model lets it run offline. The README says the packaged installer is Windows-only. On macOS and Linux you run it from source.

## What it does

- **Supervisor and subagents**: A supervisor delegates to specialist subagents with isolated context. Code changes go to a git branch and wait for your approval before merging.
- **Fail-closed permissions and vault**: Each tool is set to allow, ask or deny, and the shell stays locked down. API keys and tokens live in an encrypted vault that agents never see.
- **Layered, visible memory**: Memory has session, project, user and org layers. You can bind an Obsidian vault, Notion, a folder or an MCP memory server per project, and view it as a 3D graph.
- **Workflow canvas and scheduling**: Drag step nodes into a workflow and run it. Agents can build workflows too. Scheduled tasks use repeat presets or a set date and time.
- **Local models and phone chat**: Point it at any OpenAI-compatible local endpoint, and make a Telegram bot two-way to chat with the same agent from your phone, behind a sender allowlist.

## What you could build

- Delegate a coding task to subagents and review the resulting git diff before merging.
- Run an AI assistant fully on your own GPUs using Ollama or vLLM.
- Build a visual workflow that chains agent steps and runs on a schedule.
- Redact personal data from PDFs and Office files while keeping the original format.
- Keep separate per-client memory by binding a notes folder to each project.

## Who it's for

Suits developers and power users who want agents with auditable, approval-gated actions on their own machine. Windows users get an installer, and others need to be comfortable running from source.

## Get started

On Windows, download Iron-Jarvis-Setup-<version>.exe from the Releases page and run it. Click More info, then Run anyway, on the SmartScreen warning, since the app is not code-signed yet. On macOS or Linux, the README mentions running the daemon and dashboard from source, but the excerpt gives no full steps.



Tags: agents, orchestration, local-first, desktop-app, memory