zergrouter.com / zrouter.io

One routing plane for every agent call.

ZergRouter separates model routing from the main ZergAI app: provider credentials, quotas, usage logs, and compatibility endpoints live in a focused service that Zerg, ZTC, and ZDE can share.

View endpoints Manage providers

Route preview

OpenAI-compatible request

registry.fly.io/zergrouter:deployment-01KSVG05KNEVASN1YK7BM0GPX3

curl https://zrouter.io/v1/chat/completions \
  -H "Authorization: Bearer $ZERGROUTER_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "provider": "openai",
    "model": "gpt-5.4-mini",
    "messages": [
      { "role": "user", "content": "ship it" }
    ]
  }'

JWT

short-lived auth

AES

secret storage

SSE

stream passthrough

surface area

Small API, strict ownership.

The public router contract is intentionally narrow. ZergAI handles accounts; ZergRouter handles model traffic and provider operational state.

POSTAnthropic Messages

/v1/messages

POSTOpenAI Chat

/v1/chat/completions

POSTOpenAI Responses

/v1/responses

POSTEmbeddings

/v1/embeddings

GETModel Catalog

/v1/models

POSTZerg legacy chat

/llm/chat

Account auth stays in ZergAI

ZergAI remains the identity authority and mints short-lived router JWTs for workspaces and zerg service tokens.

Secrets live with the router

Workspace provider credentials are encrypted in ZergRouter, with platform environment keys as controlled fallback.

Provider APIs stay compatible

Use OpenAI-compatible, Anthropic-compatible, and legacy Zerg LLM endpoints without changing client ergonomics.

Usage is first-class

Every non-streaming route records model, provider, workspace, source, and token counts for quota and admin visibility.