Secure orchestration for AI agents

Launch, mesh, and monitor AI agents in real time — on your hardware, under your control.

Arkavo Edge

Instant, secure orchestration for AI agents. No configuration files, no setup — agents auto-discover via mDNS and form a mesh.

Zero config

Just run arkavo. Auto-naming, auto-routing, auto-discovery. Launch a web UI with arkavo ui.

Fast

≤ 2ms agent-to-agent latency on commodity hardware — from workstations down to a Raspberry Pi 5.

Visual

See live agent communication flows in real time and monitor your entire mesh at a glance.

Multi-provider routing

Cost-aware model selection across OpenAI, Anthropic, Gemini, Kimi, DeepSeek, and local models via llama.cpp.

Local edge models

Run Ministral 3 (3B/8B/14B) and other models locally, with vision support — your data never leaves your device.

Built-in security

Security scanning (Semgrep, OSV, SBOM) and fine-grained access control for MCP tools via OpenTDF.

Install

macOS and Linux, via Homebrew:

$ brew tap arkavo-org/homebrew-arkavo
$ brew install arkavo

Then start an agent — that's it:

$ arkavo        # start an agent (zero config)
$ arkavo ui     # or launch the web UI

Installers for macOS, Windows, and Raspberry Pi 5 (ARM64) are on the releases page.

Open Source

AI infrastructure we build in the open at github.com/arkavo-ai.

arkavo-edge

Instant, secure orchestration for AI agents — launch, mesh, and monitor in real time.

Rust

game-rl

Multi-agent AI infrastructure for games. Turn any game into a multi-agent environment — train RL policies, orchestrate LLM-powered NPCs, or build AI game masters.

a2a-swift

Swift SDK for the Agent2Agent (A2A) Protocol.

Swift

a2a-rs

Rust SDK for the Agent2Agent (A2A) Protocol.

Rust

a2a-conformance

Pairwise interop matrix for A2A protocol SDKs — N clients × M servers, scripted scenarios, wire-evidence reports.

Swift

torg

TØR-G: Token-Only Reasoner (Graph) — a zero-parser, boolean-circuit IR for formally verifiable AI agent policy synthesis.

Rust

WeDLM-rs

Rust implementation of Tencent's WeDLM parallel decoding architecture for fast LLM inference.

Rust

mlx-swift-lm

LLMs and VLMs on Apple silicon with MLX Swift.

Swift

View all repositories →