AI at EasyMeta

We build the infrastructure that makes AI safe and useful.

AI sits at the core of what we build, both inside our apps and as infrastructure others can build on. From an open-source LLM gateway to governed AI agents, we ship AI systems held to a production standard.

Safety is a design requirement

Governance, auditing, and isolation go into our AI systems from the first commit. We treat trustworthy behaviour as a requirement, not a later patch.

Own your stack

Our infrastructure is self-hostable and provider-agnostic. There is no vendor lock-in and no black box, so you keep full control of your models, keys, and data.

Engineered to last

Strict typing, high test coverage, and enforced architectural boundaries. We hold our AI infrastructure to the same standard as any production system.

What we're building

Open infrastructure, governed agents, and a safety layer for the AI ecosystem.

Helm API

Open source · MIT

An open-source LLM gateway with a single, unified API surface.

Helm sits between your application and the model providers. Point any OpenAI, Anthropic, or Gemini client at it, and Helm classifies each request, routes it through a configurable lane, and handles cross-provider failover. Every routing decision is recorded for full observability.

  • One OpenAI-compatible endpoint for OpenAI, Anthropic, and Gemini
  • Automatic provider failover with circuit breaking
  • Per-key authentication, rate limiting, and budget controls
  • SvelteKit admin console and OpenAPI-documented API
  • 95% test coverage, distributed as a self-hostable Docker image

AgentCrew

In development

A governed multi-agent runtime where AI agents work like a team.

AgentCrew coordinates role-based AI agents (a coordinator, an engineer, a reviewer) that advance projects together inside a group chat. Policy is enforced by the runtime rather than the prompt, and every action is recorded in an append-only audit log. Currently in active development.

  • Role-based agents collaborating in a shared conversation
  • Runtime-enforced governance with fail-closed policy gates
  • Append-only audit log for every agent action
  • Prompt-injection isolation for untrusted content
  • Self-evolving skills with a capture, propose, and approve lifecycle

Skillstore

Live

An independent security-audit layer for AI agent skills.

Skillstore catalogues and distributes agent skills that install into Claude, Codex, and Claude Code. Every skill is independently reviewed, version-pinned, and cryptographically signed, so teams can adopt new agent capabilities without inheriting unknown risk.

  • Independent security review of every published skill
  • Version-pinned and signed for verifiable supply-chain trust
  • Compatible with Claude, Codex, and Claude Code
  • A neutral, cross-vendor safety layer for AI agents

AI in our products

The same AI, closer to people.

The same infrastructure powers our consumer apps: AI-driven English learning, on-device intelligence, and tools that stay private by design. Solid engineering underneath, refined products on top.

Let's build together

We're an AI-native team that ships fast and thinks long-term. If you're a partner, a team evaluating AI infrastructure, or an investor who shares our view of where this is heading, we'd welcome a conversation.