ARC V2

A persistent, agent-based AI system for Linux.

Local inference · service-oriented

---

Overview

ARC V2 is a Linux-native AI system designed as a persistent environment for an agent, rather than a conventional request/response chatbot.

Its architecture separates system boot, service supervision, agent orchestration, model inference, and system capabilities into distinct components.

The long-term goal is an assistant that can remember, plan, act, react to events, and operate proactively within its environment.

flowchart LR
    Boot["Boot"] --> Pulse["Pulse"]

    Pulse --> Kernel["Kernel"]
    Pulse --> Runtime["Runtime"]
    Pulse --> Services["Services"]

    Kernel --> Runtime
    Kernel --> Services

    Runtime --> Models["LLM Models"]
    Services --> Capabilities["System Capabilities"]

Loading

Architecture

Component	Responsibility
Boot	Initializes ARC and starts the system
Pulse	Supervises service lifecycle, dependencies, readiness, health and recovery
Kernel	Agent orchestration, context, planning, memory and task coordination
Runtime	Model loading, inference and LLM execution
Services	Independent capabilities exposed to the ARC system

Note

Kernel and Runtime are Services themselves. Pulse manages them through the same service lifecycle used by other ARC components.

Core principle

The Kernel decides what should happen. The Runtime performs model inference. Services provide capabilities. Pulse keeps everything alive.

This separation allows ARC to evolve without coupling the agent's reasoning system to a specific model backend or system capability.

Current Status

Warning

ARC V2 is in early active development. The boot and service foundation are functional, while the higher-level agent system is still under development.

Area	Status
Boot foundation	✅ Complete
Service architecture	✅ Complete
Service lifecycle & dependencies	✅ Complete
Pulse supervision	✅ Complete
Runtime	🔨 In Progress
Kernel orchestration	🔨 In Progress
Persistent memory	📋 Planned
Planning system	📋 Planned
Proactive behavior	📋 Planned

Installation

Development

Clone the repository and synchronize the project with uv:

git clone https://github.com/PauWol/ARC-V2.git
cd ARC-V2

uv sync

Run ARC:

uv run arc

or for dev without rebuilding

uv run python -m arc.boot.boot

Run tests:

uv run pytest

Warning

Tests are planned but not really enrolled.

As a CLI Tool

Install ARC directly from GitHub:

uv tool install git+https://github.com/PauWol/ARC-V2.git

Then run:

arc

Update the installed version:

uv tool upgrade arc

For complete uv installation, project and tool documentation, see the uv documentation.

Configuration

ARC uses environment-based configuration.

Configuration is documented in:

docs/CONSTANTS.md

Example:

ARC_DIR=~/arc
AGENT_WORKSPACE=~/arc/workspace
LLM_MODEL_STORE=~/arc/models

Important

Do not commit secrets or private configuration files to the repository.

Documentation

ARC uses concept documents to describe the architecture and individual parts of the system.

docs/CONCEPTS.md defines the documentation standard used when creating concept pages. It describes how concepts should be structured so they remain useful to both humans and LLMs.

Document	Purpose
docs/CONCEPTS.md	Standard for ARC concept documentation
docs/SERVICES.md	Service architecture and implementation
docs/CONSTANTS.md	Environment variables and configuration

Additional concept documentation will be added as the corresponding ARC components mature.

Contributing

ARC V2 is still establishing its architecture.

For significant architectural changes, open an issue first to discuss the direction. Documentation, tests and focused improvements are welcome.

License

Warning

ARC V2 does not currently have a finalized open-source license. Until a LICENSE file is added, the repository should be treated as all rights reserved.

---

ARC V2 Building the foundation for a persistent AI agent.