Infrastructure for AI agents.
We build the systems layer between intelligent agents and the world — the primitives that make agents fast, private, and production-ready.
AI agents are becoming the primary interface between humans and software. The infrastructure they run on — the browsers they use to see the web, the memory systems they use to learn across sessions, the protocols they use to act in the world — was not designed for them. It was designed for humans.
We build that infrastructure from scratch. Every system we ship starts from a single question: what does an agent actually need, and what is just legacy from the human-first web?
Native Rust browser engine purpose-built for AI agents.
Phantom is not a Chrome wrapper. It is not Playwright with a different API. It is a browser engine written from scratch in Rust, designed around a single constraint: agents do not have eyes, they need structure.
Phantom fetches pages, parses HTML, computes CSS and layout, and serializes the result into CCT — a compact, pipe-delimited format that uses approximately 20 tokens per node versus 121 for raw JSON. It speaks the Model Context Protocol natively. It uses rquest with BoringSSL for TLS fingerprint spoofing indistinguishable from a real Chrome 130 browser.
Early Alpha — core pipeline working, interaction tools in progress.
Start from first principles. Every component we build starts with the question of what the right design is for the agent use case — not what the existing solution is and how to wrap it.
Honesty about state. We document what works and what does not. Our READMEs reflect the actual code, not the intended code. Early alpha means early alpha.
Rust by default. Memory safety, zero-cost abstractions, single binary output, and no runtime dependencies. The systems layer for AI infrastructure should be built on the most reliable systems language available.
Open source first. The primitives that agents run on should be open, auditable, and community-owned. We build in public.
| Layer | Technology |
|---|---|
| Language | Rust 1.80+ |
| Async runtime | Tokio |
| HTTP server | Axum |
| HTML parsing | html5ever |
| Layout engine | Taffy |
| JS engine | QuickJS (V8 planned) |
| Network | rquest + BoringSSL |
| Protocol | Model Context Protocol (MCP) |
| Observability | Prometheus + Grafana |
# Run Phantom Engine
docker run -d \
-p 8080:8080 \
-e PHANTOM_API_KEYS=your-key \
polymit/phantom-engine:v0.2.0
# Verify
curl http://localhost:8080/healthFull documentation at phantom-engine.
We welcome contributions from engineers who care about agent infrastructure. Read the contribution guide before opening a pull request.
Good first contributions: bug fixes with reproduction steps, test coverage, performance improvements with benchmark evidence.