Gateway features. On every key, at catalog prices.
Web search and code tools in one call, PDFs for any model, auto router, response cache, guardrails, MCP gateway, embeddings, Whisper, batches and session logs.
Eight groups, one key
Every model on the keyAgent tools
Web search, web fetch, remote MCP tools and a bash and JavaScript sandbox, run by the gateway inside one request, up to 8 tool rounds.
Files and redaction
Documents read as text for models that cannot take files; per-key secret and PII redaction, restored in the answer.
Routing
model "auto" across three tiers, fallback lists of up to 5 models, saved presets called as @preset/name, and routing by price, latency or throughput.
Caching and context
An opt-in response cache billed at a tenth of the original charge, and middle-out compression for prompts longer than the context window.
Guardrails and app sign-in
Opt-in blocked patterns, prompt-injection checks and tool policies per key or org; OAuth PKCE sign-in that mints a key for your app’s user.
MCP gateway
Register MCP servers with sealed headers and allowlists; every tool on one /mcp/org endpoint, filtered by the key’s tool policy and logged.
Endpoints
POST /v1/embeddings, /v1/moderations, /v1/rerank, /v1/audio/transcriptions and translations, /v1/images/edits, and the Batch API through /v1/files and /v1/batches.
Observability
Session timelines from x-session-id, a usage CSV, log export to a webhook, Langfuse or any OTLP backend, per-session caps and a loop guard.
Prices · live catalog, 2026-11-04
| Feature | Price |
|---|---|
| Web search | $0.01 per search |
| Web fetch | $0.001 per fetch |
| Code sandbox | $0.05 per container-hour |
| Remote MCP tool call | $0.00 per call |
| Document reading | $0.00 per page |
| Embeddings, from Text Embedding 3 Small | $0.027089 per million input tokens |
| Rerank, from BGE Reranker Base | $0.007775 per million input tokens |
| Transcription, from Whisper Large v3 Turbo | $0.0012825 per audio minute |