AI Gateway
One AI gateway for every model.
Every request on the record.
Call 90+ models from OpenAI, Anthropic, Google, xAI, and more through one API key. Set a fallback model for when a provider fails, and get a log of every request with its tokens, cost, and latency.
Free Developer plan · No card required · 90+ models, 17 providers
01
HOW IT WORKS
One key in. Any model out. Every call logged.
An AI gateway, also called an LLM gateway, sits between your app and the model providers. Your code calls one Fetch Hive endpoint, and Fetch Hive handles the provider call, the fallback, and the record of what happened.
One API key
01
Create a workspace API key for production, staging, or development. It is the only credential your app needs, whichever model answers.
A deployment
02
Pick the model and its settings in a prompt deployment. Your code calls the deployment by name, so you can change the model without shipping code.
Model fallback
03
Give the deployment a fallback model from another provider. If the call to the primary model fails, the same request goes to the fallback.
A logged request
04
Each call is saved with its provider, model, tokens, cost, duration, and status, and returns a request ID you can look up later.
02
MODELS
90+ models from the providers you already use.
OpenAI, Anthropic, Google, xAI, Perplexity, Meta Llama, DeepSeek, Qwen, and more, behind one API. Compare models by context window, vision, reasoning, tool calling, and JSON output before you pick one.
94
MODELS
Live in the catalog today, with new releases added as they ship
17
PROVIDERS
From OpenAI, Anthropic, and Google to open-weight labs
0%
MARKUP PER CALL
Hosted usage is charged at the recorded provider cost
Hosted keys
Start without a provider account. Top up your LLM wallet and each call is charged at the recorded provider cost, with no markup per call.
Your own keys
On paid plans, add your own API keys for OpenAI, Anthropic, xAI, Perplexity, Cohere, and Kimi, and those calls are billed by the provider.
Compare models
Run the same prompt against several models side by side in the app, then choose on the output, the speed, and the cost.
Model library
Every model has a public page with its context window and capabilities. The same list is available from the models API.
03
ROUTING
Route by deployment. Fall back when a provider fails.
Routing in Fetch Hive is explicit. A deployment says which model answers, a variant lets you keep a second configuration next to it, and a fallback model takes over when the primary call fails. You always know which model handled a request.
Deployments
Your app calls a deployment, not a model name. Change the model, the prompt, or the settings in Fetch Hive and the next request uses them.
Variants
Keep more than one configuration under a deployment, such as a cheaper model for a second use case, and name the variant in the request.
Fallback model
Set a second model with its own provider, temperature, token limit, and key type. It is tried when the call to the primary model fails.
Versions
Every change is saved to version history. Compare versions and roll back if a new model or prompt does worse.
prompt invoke · support-reply
# one endpoint, whichever model the deployment points at
$ curl https://api.fetchhive.com/v1/prompt/invoke \
-H ‘Authorization: Bearer <api key>‘ \
-d ‘{ “deployment”: “support-reply”, “variant”: “default”,
“inputs”: { “question”: “Where is my order?” } }’
✓ 200 OK · request_id req_8f2c41
Status
success
Model
returned with the response, so you know which one answered
Usage
token counts and cost returned with every response
Lookup
GET /v1/public/requests/req_8f2c41
HOSTED OR YOUR OWN KEYS
Each model in a deployment runs on Fetch Hive hosted keys or on your own provider key. You can mix them, for example your own OpenAI key for the primary model and a hosted model as the fallback.
Hosted usage is charged at provider cost from your LLM wallet
Wallet balance does not expire
Own keys work with six providers on paid plans
Each model call counts as one task on your plan
One response format, whichever model answers.
The same JSON fields from every provider
The same stream events: reasoning, response, usage, error
Change the model without touching your parsing code
Providers each answer in their own shape. Fetch Hive returns one, streaming or not, so moving a deployment from OpenAI to MiniMax changes nothing in your app.
04
OBSERVE
LLM observability, without adding another tool.
Tracing, logging, and cost tracking are part of the gateway. Every request is stored with what was sent, what came back, and what it cost, so you can debug a bad answer or explain a bill.
Find out why an answer was wrong, or why a bill was high.
What was sent and what came back
Which model answered, and how fast
How many tokens it used and what it cost
Every request is recorded. Filter by your own metadata to find one, then share it by link.
05
COST AND USERS
Know what each model, and each user, costs you.
The analytics dashboard adds the requests up: cost by model, token usage, latency percentiles, error rate, and cache savings. Pass a user ID with a request and the same numbers are split per user.
Analytics
COST
See spend and model health in one place.
Model costs, token usage by type, latency percentiles, and error rate for every model you call, over the date range you choose.
Total cost split by model and by service
Latency percentiles and error rate per model
Prompt cache hit rate and what it saved
Free Developer plan
User tracking
USERS
Attribute usage to your own customers.
Send a user ID with each request. The Users page shows requests, tokens, and cost for every one of them, so you can price your plans or spot heavy use early.
One optional field on the request
Top users by requests, tokens, and cost
Filter logs by any metadata you send
Works with every model
PART OF THE FETCH HIVE PLATFORM
The same gateway runs your prompts, agents, and workflows.
Agents and workflow steps call models through the same catalog, keys, and logs. Start with one prompt deployment and grow into multi-step work without redoing your provider setup.
Run prompts on any model right inside Claude
Connect Fetch Hive once, then run prompts, agents, and workflows from the chat you already have open. Free for 72 hours as a guest, no account needed.
Which AI assistant do you use?
Open Connectors
In Claude (web or Desktop), open Settings, then Connectors, and click Add custom connector.
Paste the connector URL
Name it Fetch Hive and paste this URL.
https://connect.fetchhive.com/mcpConnect as a guest
Click Connect. When Fetch Hive opens in your browser, choose Continue as guest (or sign in), then click Allow.
Then ask Claude
Run my support-reply prompt on two different models and show me the answers side by side with what each one cost.
Create a free account any time to keep what your assistant built. Stuck? Read the setup guide.
FAQ
Questions developers ask about the AI gateway.
Everything else is in the docs.
An AI gateway is a layer between your application and AI model providers. Your code calls one API, and the gateway holds the provider keys, sends the request to the right model, falls back if that model fails, and records the tokens, cost, and latency. Fetch Hive is an AI gateway with prompts, agents, and workflows built on top.
