Getting started with Max Router
Choose a model, create an API key and make your first request through Max Router.

Choose a model and endpoint
Open the model catalog and compare the available models, supported endpoints and current prices. Select a model whose input and output match your application. Keep its exact model ID and endpoint handy; capabilities can vary between models and protocols.
When an integration fails, the hard part is often finding which layer failed. Pick a model that supports a conversation endpoint and ask one short question. Leave image input, tools, streaming and elaborate system prompts for later. A small request lets you check the address, key, model ID and request format separately before adding the features your product needs.
Use the catalog to find the exact model ID, supported endpoints and current pricing. A friendly display name may differ from the identifier expected in code; do not guess punctuation or version numbers. A text endpoint is not automatically suitable for an image or video task. Recheck the documented capabilities and important parameters whenever you change models.
Create an API key
Sign in, open the API keys page and create a key for your application. Store it in an environment variable or a secrets manager. Never place a real key in browser code, a public repository or a screenshot.
Use different credentials for development and production so you can limit and investigate each environment independently. Read the key on your server from an environment variable or secrets manager. Frontend bundles, browser extensions and public repositories are not secret stores, even when a build variable contains the word SECRET. Rotate a key exposed in a screenshot or log and check its usage.
An application connects through a gateway to text, image and video models.Send a request
The example below uses the chat completions endpoint. Replace YOUR_MODEL_ID with a model that supports this endpoint and set MAX_ROUTER_API_KEY in your environment. For image and video generation, choose the endpoint shown in the model documentation instead.
curl https://maxrouterai.com/v1/chat/completions \
-H "Authorization: Bearer $MAX_ROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"YOUR_MODEL_ID","messages":[{"role":"user","content":"Hello!"}]}'
Review the result
Check the response and any reported error, then review usage in your dashboard. Start with a small request, verify the model and parameters you need, and consult the documentation before adding streaming or tools.
If the first request fails
Check the model ID and endpoint exactly as listed in the catalog. An authentication error calls for a key or permissions check; an unsupported parameter needs a change to the request. Avoid pasting a key into a public support message. A request identifier and a sanitized error are usually more useful.
A successful HTTP status is only the start: inspect the body, required fields and any completion state. For failures, retain a request ID and a sanitized error. Authentication problems call for a key and permissions check; invalid parameters require a request change. Rate limits and temporary failures may warrant a bounded retry according to the endpoint documentation, not an endless loop.
Break a request into three stages
When someone asks a question in your app, send it from the browser to your own server first. Authenticate the user and validate the input there, then call the model using a key stored on the server. Return only the result the user needs. A person inspecting browser traffic should never receive your platform key.
Trace failures along that same path: did the browser reach your server, did the server send the request, and did the API respond? A request ID, status and duration are often more useful than logging an entire conversation. Compare a minimal command-line request with the same request from your application; differences in environment variables or JSON serialization can reveal the problem.
Find the cause in the response
A successful HTTP status is the start of the check, not the end. Locate the returned content and decide how your app handles an empty response, a finish reason or a model refusal. Response shapes differ by protocol, so don't assume that a field path from one example works everywhere.
For authentication failures, check the key, its environment and permissions, and the authorization header. For invalid parameters, strip the request back to its required fields and add options one at a time. Follow documented guidance for temporary rate limits. When asking for help, share the request ID, time and a redacted error—not a key, private prompt or customer file. Rotate a key if you suspect it has leaked.
Run a small pilot before launch
Before launch, try a short question, a long input, a missing field, an invalid key and a cancelled request. Check loading, completion and error states, server timeouts and repeated clicks. Use separate test credentials and examples without real customer data.
Invite a small group to try realistic tasks and record quality, duration, usage and failures. Expand gradually rather than automatically retrying every error. Keep a record of the model ID, endpoint, dependency versions and test date; rerun the same samples after changes. If documentation and observed behavior differ, contact support with a minimal redacted request and its ID.
Take the next step carefully
Once a small request works, try a representative input and compare the result with what your product actually needs. Add features one at a time, keep credentials on your server, and review usage before increasing traffic. The model documentation is the source for its current limits and supported options.
Try representative examples without real sensitive data and record quality, latency, usage and errors. Exercise an invalid key, interrupted network and user cancellation so the interface can explain each outcome. Increase traffic gradually after the basic workflow works. Model capabilities, prices and rate limits can change, so check the current documentation again before a larger launch.
Record the model ID, endpoint, non-sensitive settings and date of the first successful test. A small reference case makes later changes to dependencies or models easier to compare. Do not save keys or complete customer inputs in the note. When asking for help, a minimal request shape, request ID and sanitized error are safer and more useful than an entire production log.