Open source · MIT licensed · v3

Route every message
to the right AI agent.

Self-hosted platform connecting Discord, Telegram, Slack, WhatsApp — and any messaging app — to teams of specialized AI agents. Silent failover. Automatic memory. Full cost tracking.

Star on GitHub Read the docs →

npx arvis

>_<arvisv3self-hosted AI agent platform
✓Core started
✓Discord connectoronline
✓Telegram connectoronline
✓4 agents loadedsupport · analyst · coder · conductor
✓Dashboard readyhttp://localhost:5100

8

Platforms

7

LLM providers

MIT

Licensed

100%

Self-hosted

TS

TypeScript

0

Lock-in

★ Unique feature

No API bills.
Run on your Claude subscription.

Arvis runs agents through the Claude CLI — meaning your Claude Pro ($20/mo) or Max ($100/mo) subscription covers your entire agent fleet. No per-token costs. No surprise bills. Stack multiple accounts for even more throughput.

API Keys

✗$0.015 per 1K tokens

✗Bills spike with usage

✗Budget surprises

✗Rate limits hit hard

Subscription

✓Flat $20–$100/mo

✓Zero per-token cost

✓Predictable spend

✓Stack accounts freely

Works alongside API keys — Arvis silently rotates between subscription accounts and API keys based on priority and rate limits.

.env

# Account 1 — Claude ProCLAUDE_CLI_HOME=/home/arvis/acc1# Account 2 — Claude MaxCLAUDE_CLI_HOME_1=/home/arvis/acc2# Account 3 — API key fallbackANTHROPIC_API_KEY=sk-ant-...# Model to use for CLI accountsCLAUDE_CLI_MODEL=claude-sonnet-4-6

3 accounts loaded — 2 subscription, 1 API key

One-time setup: Run HOME=/home/arvis/acc1 claude login for each account. Arvis handles rotation automatically.

Features

Everything an agent platform needs.

Arvis handles the hard parts — routing, memory, failover, cost, queuing — so you just define agents and connect platforms.

Multi-platform routing

Discord, Telegram, Slack, WhatsApp, Matrix, Web, SMS, Email — every message lands at the right agent automatically.

Silent failover

Rate limit hit on Anthropic? Arvis silently switches to your next account. Users never see an error. Ever.

Conversation memory

Sticky facts persist across sessions. Context is auto-compacted when it grows large. Every agent remembers what matters.

Cost tracking

Per-agent, per-provider, per-model cost breakdown. Know exactly what each conversation costs across all your LLM accounts.

Job queue + scheduler

Priority queue with automatic retries. Cron and heartbeat workflows. Stuck job recovery. Everything visible in the dashboard.

Agent delegation

Agents spawn sub-agents for parallel work using simple DELEGATE markers. Build pipelines without writing orchestration code.

Why Arvis

Stop fighting your bots.
Start shipping agents.

Without Arvis

✗Rate limit hits → users see 500 errors

✗Bot crashes → conversation history wiped

✗N bots → N dashboards to manage

✗No idea what your agents are doing

✗Context limits hit → quality drops silently

✗New provider = rewrite your whole stack

With Arvis

✓Rate limit → silently switch account

✓Job queue → retry with full history

✓All bots → one dashboard

✓Real-time queue monitor + logs

✓Auto-compaction keeps context fresh

✓Add provider = one new config entry

How it works

Up and running in 4 steps.

01

Connect your platforms

Add Discord, Telegram, Slack, or WhatsApp tokens in .env. Connectors start automatically.

02

Define your agents

Create specialist agents — support, analyst, coder, creative. Each with their own system prompt, model, and tools.

03

Messages route automatically

Arvis reads each message and picks the right agent based on platform, channel, keyword, or agent rules.

04

Monitor everything

Real-time dashboard shows every job, conversation, cost, and log. No black boxes.

LLM Providers

Every major provider.

Use your Claude subscription (zero per-token cost) or any API key. Multiple accounts per provider — Arvis rotates silently on rate limits.

★ No API cost

Claude CLI

Pro & Max subscription — no API cost

Anthropic

claude-haiku/sonnet/opus

OpenAI

gpt-4.1, gpt-4.1-mini, o4

Google

gemini-2.5 pro + flash

Ollama

any local model, zero cost

OpenRouter

100+ models, one key

Custom API

any OpenAI-compatible URL

Platforms

Every messaging platform.

One agent stack, every platform. Add connectors by setting env vars — no code required.

Discord

Slack

Matrix

Web

SMS

Quick start

Running in under 5 minutes.

Node 20+, one API key or a Claude subscription. Done.

1. Clone and install

git clone https://github.com/Arvis-agent/arvis
cd arvis && npm install

2. Configure (pick one)

cp .env.example .env

# Option A — Claude subscription (no API cost):
HOME=/opt/arvis/acc1 claude login
CLAUDE_CLI_HOME=/opt/arvis/acc1

# Option B — API key:
ANTHROPIC_API_KEY=sk-ant-...

3. Start

npm start
# Dashboard → http://localhost:5100

4. Or Docker (easiest for VPS)

docker-compose up -d
# Core + dashboard start together

Read full setup guide →Deployment guide →

★Open source on GitHub — MIT licensed

Self-hosted.
No lock-in. Runs anywhere.

Your agents, your servers, your data. No subscription required.

Star on GitHub Get started →

Get release notifications

New versions, features, and important updates straight to your inbox.

Route every messageto the right AI agent.

No API bills.Run on your Claude subscription.