Everything your agents run on.
The tools an agent calls, the routing it runs on, the rules that hold it and the traces that explain it. One key, one prepaid balance, every price read live.
Feature pages
Web search, web fetch, remote MCP and a code sandbox run inside one API call, in the OpenAI and Anthropic tool shapes, priced per call.
PDFs and office files read by any model, and secrets or personal data replaced before the prompt leaves the gateway.
Relayswitch/auto picks a model per request, fallback lists, saved presets, throughput and latency routing.
An opt-in response cache billed at a tenth, and middle-out compression for prompts longer than the context window.
Blocked patterns, a prompt-injection check and tool allow and deny lists, plus Sign in with RelaySwitch for apps.
Register your MCP servers once and give every agent one MCP endpoint on its API key.
Embeddings, moderation, rerank, transcription, image edits and the Batch API on the same key.
A prepaid budget, a daily, weekly or monthly spend limit, RPM, TPM and concurrency caps, an allowed-model list and an expiry on every key, so a runaway agent stops at its own line.
Sessions, a timeline per agent run, log export to Langfuse, Datadog, OTLP or a webhook, session caps and a loop guard.
Also on the key
- Hosted agents: a model, webhook skills, MCP servers, memory and a schedule, run for you
- Workflows: a JSON graph of model calls, run by one request
- Signed receipts: an Ed25519 receipt for every request, checked without an account
- OpenAPI document: every route, field and header, OpenAPI 3.1
- MCP server: https://api.relayswitch.io/mcp: models, estimates, the daily check and chat as MCP tools
- Wire formats: OpenAI Chat Completions and Responses on 326 models, Anthropic Messages on 326, Gemini generateContent on 326
One key for the whole agent stack
Top up from $10 by card or crypto, create a key per agent, turn on what it needs.