Eigin Relay

Eigin Relay is a connectivity service for Eigin agents. It brokers the few capabilities an agent can't perform on device because they need a server-side key or a stable endpoint: today, web search and AI inference. The relay cares about valid credit, not who you are: show proof of purchase and you can use it, no account needed.

Web search queries go to the relay, which forwards them to Staan, a European search API operated by the Qwant + Ecosia joint venture, EU-hosted in France with zero data retention, and returns the results with LLM-ready content snippets. Search draws from a shared usage pool; when it's exhausted, the app falls back to on-device search.

Inference

The relay forwards inference to a vetted upstream provider, so users can run capable models without their own key. Prompt and response content is never logged or stored; the relay reads only the token counts it needs to compute cost.

Inference is gated by credits, tied to a purchase or grant rather than an identity:

  • Users buy non-expiring credit packs via IAP, or receive a direct grant. Either creates a wallet identified by a hash of the transaction, not the person.
  • A request is admitted when the wallet has balance. Credits are debited after completion, based on actual upstream token usage. Roughly 1 credit buys €0.01 of inference at upstream cost. Credit packs are priced at their credit value, plus the store's platform fee and our margin.

Model assets

On-device models (the local LLM and the TTS voices) download from a mirror the relay fronts, not from their original HuggingFace repos. The relay serves a versioned manifest per asset — file list, SHA-256 checksums, and short-lived presigned download URLs — and the app verifies every file before use.

Abuse prevention

  • A shared key (x-eigin-key) gates every request.
  • Per-IP rate limiting caps call speed.
  • Per-purchase rate limiting caps how many devices consume one credit pack.
  • Shared quota pools bound total usage; an exhausted pool pauses the feature until it resets.

API

See the live API docs at https://relay.eigin.ai/docs.

Self-hosting

The relay ships as a Docker image with Terraform to deploy it. You can self-host to run your own instance with no shared limits.