Agentnetes
Zero to a Self-Organizing AI Agency. On Demand.
k8s orchestrates containers. a8s orchestrates AI agents.
Give it a goal in plain English. Get a self-organizing swarm of specialist AI agents running concurrently in isolated sandboxes. No hardcoded roles. Fully emergent team composition.
See It In Action
Watch the 2-Minute Demo
Watch how Agentnetes decomposes a goal, spawns specialist agents, and produces results — all in under 2 minutes.
Want to see the full architecture?
Dive into the research foundations, sandbox providers, A2A protocol integration, and the complete technical stack.
Why Agentnetes?
Just as Kubernetes brought declarative goals, parallel execution, lifecycle management, and isolation to containers — Agentnetes brings the same to AI agents.
The Problem
- Single-agent systems hit context window limits
- Hardcoded multi-agent roles are brittle
- No isolation — agents corrupt shared state
- Agents can't verify their own work
The Kubernetes Analogy
Agentnetes
- Emergent roles from goal + codebase analysis
- Parallel execution in isolated sandboxes
- Write → test → verify → repeat loop
- Fault-tolerant: one agent failing never blocks others
How It Works
Three Phases. Fully Autonomous.
From a plain-English goal to a structured report with findings and artifacts.
Plan
Fully emergent team composition
Root agent (Tech Lead) explores the repo and decomposes the goal into emergent specialist roles. No hardcoded templates.
Execute
Parallel sandboxed execution
Each specialist runs in an isolated Docker sandbox with only two tools: search() and execute(). Agents write code, run tests, and fix failures autonomously.
Synthesize
Structured findings & artifacts
Root agent reads all worker summaries and produces a structured report with findings, artifacts, and recommendations.
See the full 6-step workflow
The documentation site has an interactive breakdown of goal specification, decomposition, sandbox isolation, auto-research execution, collaboration, and artifact delivery.
Get Started in 30 Seconds
Three Steps. No Install Required.
Get a free Google API key, pull a Docker image, and run on any git repo.
Get a free API key from aistudio.google.com
Pull the Docker image (one-time)
❯docker pull node:20-alpine
Run on any git repo
❯cd your-project❯GOOGLE_API_KEY=your_key npx agentnetes run "add comprehensive test coverage"
Ready for more? Check the full setup guide.
Configuration options, 5 sandbox providers (Docker, Vercel Firecracker, E2B, Daytona, Local), model selection, web UI setup, and environment variables.
What Can You Build?
Tell Agentnetes what you want in plain English. The swarm figures out the rest.
"Add comprehensive test coverage for all utility functions""Implement dark mode across the entire application""Run a security audit and report all vulnerabilities""Refactor the authentication module to use JWT tokens""Add TypeScript types to all untyped JavaScript files""Set up CI/CD with GitHub Actions""Add @ai-sdk/deepseek with streaming support"Deep Dive
Read the full announcement for the design philosophy and the story behind Agentnetes.
Ready to Deploy Your First Swarm?
Everything you need — architecture docs, getting started guides, sandbox configuration, model selection, and a live demo — is on the Agentnetes site.