The Problem

Building production-grade voice AI is painful. You stitch together STT, LLM, and TTS providers, write glue code for telephony, build dashboards for monitoring, and end up with a fragile pipeline that breaks every time you swap a vendor. Your agents are stateless, your tools are hardcoded, and customizing anything requires a full engineering cycle.

What is Arkenos?

Arkenos is an open-source orchestration layer that gives every voice agent a personal runtime computer. Instead of building dumb pipelines, you deploy intelligent agents that can:

• Process and reason across live conversations with function calling and tool execution
• Manage persistent memory to maintain context across sessions and interactions
• Perform real-time tasks — database lookups, API calls, booking, scheduling — mid-conversation
• Connect to MCP servers for universal tool access without custom integrations
• Self-customize through on-platform coding agents that modify agent behavior via natural language

You define what your agent should do. Arkenos handles the infrastructure.

Key Capabilities

Composable Voice Pipeline

Swap any component without rewriting your stack. Plug in your preferred STT, LLM, and TTS providers and Arkenos orchestrates the full real-time audio pipeline with sub-second latency.

Runtime Agent Computer

Every agent gets an isolated execution environment at runtime. Agents can call functions, hit HTTP endpoints, trigger webhooks, and execute multi-step workflows — all while maintaining a live voice conversation.

On-Platform Customization

Configure every aspect of your agent through the dashboard or let on-platform coding agents handle it. System prompts, voice selection, STT provider, function definitions, webhook chains — all configurable without touching infrastructure code.

Enterprise Telephony

Full telephony stack out of the box. Inbound routing via SIP trunks, outbound dialing, warm and cold call transfers, phone number provisioning and management — all through the API or dashboard.

Call Intelligence

Every conversation is automatically analyzed post-call:

• AI-generated summaries and action items
• Sentiment analysis with confidence scoring
• Topic extraction and outcome classification
• Full transcript with speaker labels and timestamps

Cost Observability

Track spend across every provider, every agent, every call. Per-session cost breakdowns across STT minutes, LLM tokens, and TTS characters with timeline charts and agent-level aggregation.

Architecture

arkenos/
├── frontend/          Next.js 16 · React 19 · Tailwind · Better Auth
├── backend/           FastAPI · SQLAlchemy · PostgreSQL
├── agent/             Python · LiveKit Agents SDK
└── docker-compose.yml

Caller ──→ Twilio SIP ──→ LiveKit Room ──→ Arkenos Agent
                                               │
                          ┌────────────────────┤
                          ▼                    ▼
                    STT Provider          TTS Provider
                    (AssemblyAI)          (Resemble AI)
                          │                    ▲
                          ▼                    │
                     LLM Engine ──→ Tool Execution ──→ MCP / Webhooks / APIs
                   (Gemini 2.5 Flash)    │
                                       ▼
                                  Memory & State

Quick Start

git clone https://github.com/Arkenos-World/Arkenos.git
cd Arkenos

# Configure environment
cp .env.example .env
cp backend/.env.example backend/.env
cp frontend/.env.example frontend/.env.local
# Edit each .env with your values

# Launch
docker-compose up -d --build

Open http://localhost:4200 to access the dashboard.

For local development setup, see the Development Guide.

Note: The agent service requires no .env file — it fetches all API keys from the backend dashboard on startup.

Cloud Deploy

Railway (One-Click)

The Railway template provides a true zero-configuration deployment. It automatically:

• Provisions the PostgreSQL database (arkenos-db)
• Deploys the Backend (arkenos-backend), Frontend (arkenos-frontend), and Agent (arkenos-agent) services
• Wires all internal routing, ports, and database credentials securely

1. Click the button above and click Deploy
2. Wait for all 4 services to show as "Online" (the backend will automatically run database migrations)
3. Open your deployed Frontend URL (https://arkenos-frontend-....up.railway.app)
4. Sign up for a new account on the dashboard
5. Navigate to the Settings page in your dashboard to plug in your provider API keys (LiveKit, Resemble, Twilio, Google, STT, etc.)

Render

Render deployment is also supported via the included render.yaml blueprint. See Render docs for details.

Documentation

Full documentation is available at arkenos.mintlify.app — quickstart, local development, configuration guides, architecture deep-dives, and interactive API reference.

The backend exposes 32+ REST endpoints across 8 routers covering agents, sessions, telephony, costs, usage tracking, voice management, and real-time communication. See the API Reference for the full interactive playground.

Roadmap

• Multi-LLM orchestration (Claude, GPT, Llama, etc.)
• Multi-TTS provider support (ElevenLabs, PlayHT, Cartesia)
• Persistent agent memory across sessions
• MCP server integration for universal tool access
• Visual graph builder for multi-agent flows
• On-platform coding agents for agent customization
• Embeddable voice widget for web apps
• Audio recording storage and playback
• Arkenos Cloud (managed platform)

Contributing

We welcome contributions from the community. See CONTRIBUTING.md for guidelines on how to get started, and read our Code of Conduct before participating.

License

Arkenos is open-source software licensed under the GNU Affero General Public License v3.0.

This means you can freely use, modify, and distribute Arkenos, but any modifications to the codebase must also be made available under the same license — including when running a modified version as a network service.

For commercial licensing inquiries, contact [email protected].