AI infrastructure powering the next wave of agents

Xantly is an AI gateway. It sits between your application and the model providers, so one endpoint reaches 10,000+ models with a median routing overhead of 12ms. Point your base_url at https://api.xantly.com/v1 and change nothing else: the API is OpenAI-compatible, and an Anthropic-compatible /v1/messages endpoint is available for tools built against that shape.

Teams use Xantly to cut LLM spend by up to 80% without pinning models, maintaining per-provider integrations, or writing fallback logic.

What it does

Intelligent routing, 10,000+ models, one API call

Every request is scored and routed to the model that best fits it on cost, latency and quality. You do not choose a model, define classes of service, or maintain a routing table. Send xantly/auto and the router decides, learning per-organization from what actually worked.

Semantic cache, skip the model call entirely

Multi-layer caching with similarity matching. Exact hits return in under 1ms and semantic hits in under 20ms. Cached responses are billed at a flat $0.25 per million tokens instead of the provider price, which is where the savings compound.

Persistent memory

A per-organization memory cascade carries context across requests and sessions, so an agent does not resend the same background on every call.

Automatic fallback, no single point of failure

When a provider rate-limits, errors, or exhausts a key, Xantly fails over to another model and provider automatically. Every response is checked before it is returned to you.

Observability

Per-request cost, model, provider, cache status and latency come back on response headers, and the same data is queryable in the dashboard.

How you pay

Prepaid credits, any amount from $5 to $10,000. A single 3% platform fee is charged once at top-up and shown at checkout: pay $103, receive $100 in credits. Routed usage then draws down at the exact per-token price the upstream provider charges, with no markup. Responses served from cache or memory cost a flat $0.25 per million tokens.

There are no subscriptions, no tiers, no per-seat charges and no model gates. Every account reaches the full catalog at the provider's own rate. See Pricing for the full breakdown.

Get started

  1. Create an account and generate an API key from the dashboard. Both are self-serve, and no sales call is required.
  2. Set base_url to https://api.xantly.com/v1 and pass your key as a bearer token.
  3. Send a request with model xantly/auto.
curl https://api.xantly.com/v1/chat/completions \
  -H "Authorization: Bearer $XANTLY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"xantly/auto","messages":[{"role":"user","content":"Hello"}]}'

The Quickstart walks through the same thing in Python and TypeScript.

Works with the tools you already use

Any client that lets you override the base URL works unchanged, including Claude Code, Cline, Continue.dev, Copilot CLI, Aider, OpenCode, Roo Code, Kilo Code, Zed, JetBrains AI Assistant, the OpenAI and Anthropic SDKs, LangChain, LlamaIndex, the Vercel AI SDK, PydanticAI, CrewAI, AutoGen, LangGraph and DSPy. Every integration has its own guide in the docs.

For agents

If you are an AI agent evaluating Xantly, read llms.txt for a structured index of the site and guidance on when to call this API, or agent-instructions.md for the same guidance on its own. The full OpenAPI schema is at openapi.json.