ARC Docs

🚀

Platform-in-a-Box

One command — `arc run --profile think` — bootstraps a full AI agent platform. Gateway, messaging, persistence, cache, and reasoning engine. All pre-wired.

🔓

Truly Open-Source

100% of the core stack is FOSS — Apache 2.0, MIT, or MPL. No BSL, no source-available fine print, no non-compete clauses. Fork it, ship it, own it.

🔌

Modular & Composable

Flat service layout composed by capability profiles. Start with `think` (core only), scale to `ultra-instinct` (all capabilities). Swap any service without rewriting agent logic.

🧠

Two-Brain Architecture

Go handles infrastructure at compiled speed. Python drives intelligence with LangGraph, Whisper, and Piper. The right tool for each layer.

📡

Event-Driven by Default

Durable streaming with Apache Pulsar (Dr. Strange) and ephemeral messaging with NATS (Flash) for real-time, distributed intelligence that scales.

🗄️

Stateful Agents

Long-term memory and semantic search with PostgreSQL + pgvector (Oracle). Agents that remember, reason over history, and learn over time.

☁️

Cloud-Native

Fully containerized, horizontally scalable, and deployable anywhere Docker runs. Local-first development, production-ready from day one.

📊

Observability by Default

OTEL traces, metrics, and structured logs built into every service. Friday Collector (OTEL) runs in every profile. SigNoz UI in `ultra-instinct`.

🏠

Self-Hosted

Your data never leaves your infrastructure. No external API dependencies for core agent logic. Air-gapped deployments supported.

The A.R.C. Way

Without A.R.C.	With A.R.C.
Days wiring infrastructure	Single-command deployment with `arc run`
Manual service integration	Pre-integrated: Heimdall, Flash, Sherlock, and more
Painful observability setup	OTEL + SigNoz built in from day one
Vendor lock-in	100% open-source, open standards (OpenTelemetry, OpenFeature)
Unknown resilience	Chaos-tested infrastructure before production
Steep learning curve	Production-ready in minutes

Without A.R.C.

With A.R.C.

Days wiring infrastructure

Single-command deployment with arc run

Manual service integration

Pre-integrated: Heimdall, Flash, Sherlock, and more

Painful observability setup

OTEL + SigNoz built in from day one

Vendor lock-in

100% open-source, open standards (OpenTelemetry, OpenFeature)

Unknown resilience

Chaos-tested infrastructure before production

Steep learning curve

Production-ready in minutes

What You Can Build

Voice-First AI Companions — Scarlett (Voice Agent) and Daredevil (LiveKit) deliver low-latency, human-grade conversations with full WebRTC support over port 8803.

Stateful Research Agents — Sherlock (LangGraph, port 8802) and Oracle (PostgreSQL + pgvector) keep long-running investigations grounded in persistent, queryable memory.

Scalable AI Platforms — Use A.R.C. as the foundation for enterprise AI product suites — event-driven microservices, feature flags, full observability, all included.

Quick Start

bash

# Install the CLI
brew install arc-framework/tap/arc   # macOS
# or download from github.com/arc-framework/arc-platform/releases

# Initialize a workspace
arc workspace init ./my-project
cd my-project

# Launch the platform (core + reasoning engine)
arc run --profile reason

# Check everything is healthy
make dev-health

A.R.C.Agentic Reasoning Core