Skip to content

Open source AI harness · macOS · MIT

The open-source AI harness that lives on your Mac.

G-Rump is the agent loop, 153 tools, memory, and safety gates that let a model actually do work — shipped as a native Swift app that learns from every run. Bring your own key. It’s free.

The G-Rump app on macOS — chat, agent modes, and the panel rail

Anatomy of a harness

A harness is everything between the model and your machine.

Recursive self-learning

Most agents forget every session. This one doesn’t.

  1. 01
    Actthe agent loop
  2. 02
    ObserveOutcomeLedger
  3. 03
    DistillReflectionEngine
  4. 04
    PersistLessonStore
  5. 05
    ApplySkillProposals

01 · Act · the agent loop

The agent does the work.

A run is real work — reading files, editing code, running builds and tests through 153 tools. Every run leaves a trail: which tools ran, what failed, where the agent had to pivot.

02 · Observe · OutcomeLedger

Every outcome is recorded.

The outcome ledger persists each run's signals: tool stats, build failures, loop pivots, review criticals. Success is two-stage — if your next message is a correction, the run is re-scored. Being corrected is data.

03 · Distill · ReflectionEngine

Outcomes become lessons.

A reflection pass — cheap model first — turns raw outcomes into short, imperative lessons. It can add, reinforce, weaken, or revise them, and it tells you in chat when it does.

04 · Persist · LessonStore

Lessons earn their keep.

Lessons carry a confidence score that moves with wins and losses, decays when idle, and auto-retires when it stops being right. The top lessons ride along on your next prompt — attributed, so results flow back.

05 · Apply · SkillProposals

Strong patterns become skills — with your approval.

When lessons cluster, the agent proposes a skill: a diff you review in the Learning panel. Approve it and it becomes part of the harness. Reject it and it's never proposed again. Nothing is applied silently.

Walk through the loop in the docs →

Native macOS

A first-class Mac citizen, not a process in a terminal.

0

Electron

Real SwiftUI. Real Keychain. 13 MB on disk.

20

dock panels

Build console, git, terminal, simulator, tests, logs, profiling — plus a ⌘0 navigator.

⌘R

build & run

xcodebuild or SPM, straight to a booted simulator with live app logs.

LSP

live diagnostics

SourceKit-LSP feeds real symbols and errors into the agent's context.

⌃Space

quick chat

Double-tap from any app. Spotlight, OCR, and Calendar are tools too.

Security model

This app runs LLM-directed shell commands. Here is what stands in the way.

  1. 01

    Exec approvals

    Four levels per binary — Deny (default), Ask, Allowlist, Allow — with Strict, Balanced, and Permissive presets.

  2. 02

    The Conscience gate

    A deterministic, fail-closed check before any mutating tool. It refuses destructive shell patterns, pushes to protected branches, and writes to secret paths.

  3. 03

    Protected writes

    Changes to SOUL.md, MIND.md, or skills require explicit approval — even in normal runs. The agent cannot rewrite its own identity quietly.

  4. 04

    The daemon is off by default

    When you opt in, it works one goal at a time on a scratch branch, behind the same gates, and never pushes.

  5. 05

    The MCP host refuses run_command

    Port 18790 exposes G-Rump's tools to other clients — but never your shell.

  6. 06

    Keychain-only keys

    API keys live in the macOS Keychain and requests go straight to your provider. No accounts, no telemetry middleman, no backend.

The full model, including what’s in scope for reports, lives in the security docs.

Bring your own key

It’s free.

No subscription, no credits, no account. G-Rump is MIT-licensed open source. You bring an API key from Anthropic, OpenAI, Google, or OpenRouter and pay your provider directly for exactly what you use.

AspectG-RumpClaude CodeAiderOpenHands
Runs asnative macOS appterminal CLIterminal CLIweb UI + sandbox
Written inSwiftTypeScriptPythonPython
LicenseMITproprietaryApache-2.0MIT
BYOK multi-provider4 providersAnthropic-centricyesyes
Cross-session memory built inyesproject filesnono

All four are good tools. G-Rump’s bet is that a coding agent should be a first-class Mac citizen with a memory, not a process in a terminal.

Get started

Five minutes from download to first task.

Or clone it and read every line before you run it — that’s the point of open source.

git clone github.com/Aphrodine-wq/G-Rump.git
G-Rump 2.0.012.4 MB

macOS 14 or later · Apple silicon & Intel · Free, MIT licensed

Download for macOS
SHA-256f6ba9c194a07f490018e15c8e86db81d19dc13ab0d551c3c877ab22760e7f61c