Works with the SDKs you already use
Point the Anthropic or OpenAI SDK at your gateway. Messages, chat completions, responses, token counting and model listing all work as they do upstream.
LLM gateway
One gateway for every model your company uses.
Give every team and tool a single endpoint that speaks the Anthropic and OpenAI APIs, runs on your own provider keys, and applies the same routing, privacy and budget rules to every call.
# Claude Code, on your own provider keys
export ANTHROPIC_BASE_URL=https://<host>/v1/api-groups/eng
export ANTHROPIC_AUTH_TOKEN=<access key>
claude
# Python
client = Anthropic(
base_url="https://<host>/v1/api-groups/eng",
api_key="<access key>",
)Same rules, budgets and audit for every call
One endpoint
Drop-in for the SDKs your teams already use.
Your keys, never handed out
Calls run on your own provider accounts. Callers get a gateway key or a personal access key; the provider key never leaves the platform.
One format, every provider
Callers use the format they already know, and RocketKind talks to each provider in its own.
Automatic fallback
List providers in order, and calls move to the next one when a provider is unavailable.
Any model, same rules
Switch to a new model the day it ships.
Swappable
The model behind the agents
- Choose the model per task, or let a router pick by cost or capability
- Bring your own provider keys
- Run open models inside your network, through a tunnel
- Move to a new model with a new version, not a rebuild
- Next: tests that re-run on their own before a model takes over
Stays put
Your operations and rules
- Workflows and agents
- Your data and its permissions
- APIs and MCP tools
- Approvals, budgets and audit
Routing and privacy
The right model for each call, and nothing private in the prompt.
Routing rules
Callers ask for one model name, and RocketKind picks the right model for each request by cost, size or content.
Judged routing
Let a judge model read the request and decide where it goes. A model from any connected provider can judge, including TypeSafe AI’s Jev.
Allow lists and aliases
Choose which models each connection may serve, and give models your own names so callers never hard-code a vendor’s.
Personal data stays in
Mask or block emails, phone numbers, card numbers, US Social Security numbers and IBANs before a prompt leaves, plus your own patterns and term lists.
Cost
Know what every call costs, and cap it.
Every call priced
Tokens turn into money on every call, so spend is known as it happens.
Budgets that hold
Cap tokens per minute and spend per day or month for each key, and set ceilings for each agent, workspace and organization.
A record when you need one
Keep a record of what was sent, encrypted under your key, and readable only by the people you allow.
30+ providers
Use frontier labs, open-model hosts and your own cloud accounts side by side.
- Anthropic
- OpenAI
- Azure OpenAI
- Gemini
- Vertex AI
- Amazon Bedrock
- Mistral
- Cohere
- xAI
- Groq
- Together
- Fireworks
- DeepSeek
- OpenRouter
- Ollama