One endpoint in front of every model.
Your own keys and your own self-hosted models behind a single OpenAI-compatible endpoint, with routing, failover, per-agent cost attribution, and ceilings that refuse requests instead of reporting the overrun.
56 models · 6 providers · Bring your own keys · No token fees
Swap the model without touching the agent.
The model is configuration, not code. Change it in one place and every agent routed through the gateway follows, including the ones your engineers wrote. One key can back several provider instances, so staging and production run against the same vendor, billed and budgeted apart.
When a provider degrades, the same model answers somewhere else.
Identical output, no code change, no downtime, because the model and the provider serving it are two different decisions.
The model name is the same on both rows. The agent asked for a model, not for a vendor, so failing over changes nothing about the answer it gets.
Spend attributed to the agent that caused it.
One bill per provider becomes a line per agent, per team and per model, with ceilings that refuse at the cap, not report afterwards.
You are billed by your providers directly at your own negotiated rates. Redpanda charges no token fees and takes no share of them. Cache reads are billed at roughly 0.1× the input rate.
A base-URL swap, and your keys stop being in your code.
Nothing is rewritten. Existing SDKs and coding tools point at one endpoint instead of several, and everything above becomes true of traffic that was already flowing.
Provider credentials move into the vault. Agents only ever see a short-lived virtual key.
client = OpenAI(
base_url="https://aigw.acme-prod.cloud.redpanda.com/llm/v1",
api_key=os.environ["REDPANDA_ADP_TOKEN"],
)
# The vendor SDKs work unchanged. The gateway speaks
# OpenAI Chat Completions, Responses, Anthropic
# Messages, Gemini and Bedrock SigV4.Build one, or bring your own
Assemble a declarative agent from a model, ready-made tools and a prompt, or register the agent you already wrote and have it inherit the same governance.
How it works →MCP GatewayOne path to every system
Every tool call brokered through one place: hosted connectors with auth handled, per-agent tool scoping, and field-level policy on the way back.
How it works →Swap one base URL. Keep your rates.
You will see the first request in the cost breakdown, attributed, within seconds.