Quick Start

Tapix AI is an AI model aggregation gateway. It is compatible with OpenAI's API format — with one base URL and one key, you can call models from multiple vendors (chat, image, video, audio).

What is this

You don't need to register, top up, or reconcile separately with each model vendor. Point your requests to us, and the gateway handles the rest — routing, billing, and retries.

In one sentence:Just swap the base_url; no other code changes needed. Projects already using the OpenAI SDK usually need just two lines changed to run.

Three integration essentials

Item	Description
Base URL	`https://api.tapixai.com/v1`(OpenAI-compatible path)
API Key	Generated in the console under "Token Management" and placed in the request header `Authorization: Bearer sk-fg-xxx`
Model name	Look it up in the model catalog, or call `GET /v1/models`

Use the API Key above for model requests. Account management endpoints (such as querying account information) require the system access token in the next section.

System access token

Click "Personal Settings" on the left.
Open the Account & Security tab.
Find the "API Call Token" and click "Generate Token".

After generation, it is automatically copied to the clipboard and displayed masked on the card. Click the eye icon to reveal the full key, or the copy icon to copy it again. The one at the top of the page ID: is your user ID — you can also click to copy it.

Every click of "Generate Token" immediately invalidates the old token. After refreshing the page, it won't be shown in plain text — only as a mask. If you didn't save it, just generate a new one.

Recommended learning path

First get a non-streaming call working (the curl below will do)
Then switch to streaming to confirm your client can handle SSE
Then handle async tasks like image or video (submit first, then poll for results)
Finally, review error codes and rate limits, and get your retry logic right

Chat (Chat Completions)

One OpenAI-compatible call. Pick a snippet and copy it. Replace the model id with any live catalog name.

Anthropic Messages API

Claude-family models can also be called with an Anthropic-compatible POST /v1/messages request. The endpoint is https://api.tapixai.com/v1/messages (include /v1). Tapix AI is not Anthropic and is not affiliated with them. If your client fails, use the OpenAI-compatible /v1/chat/completions path instead.

curl https://api.tapixai.com/v1/messages \
  -H "x-api-key: ***your key" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"claude-opus-5-5","max_tokens":1024,
       "messages":[{"role":"user","content":"Hello"}]}'

Streaming Output

Add to the request "stream": true, returning SSE. Each chunk looks like: data: {...}, the last line is data: [DONE].

curl -N https://api.tapixai.com/v1/chat/completions \
  -H "Authorization: Bearer ***your key" \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-opus-5-5","stream":true,
       "messages":[{"role":"user","content":"Tell me a short joke"}]}'

Image generation (coming soon)

Image generation and other per-request audio/video models are not on sale yet. The example below is for when the endpoint is enabled. Use chat completions for models that are listed as live in the catalog.

curl https://api.tapixai.com/v1/images/generations \
  -H "Authorization: Bearer ***your key" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-image-2-c","prompt":"An orange cat standing in the rain","size":"1024x1024"}'

Long-running tasks like video and music use 'async tasks': submit first to get a task ID, then poll GET /v1/tasks/{id}. Such tasks are billed per request or per second.

Model list

curl https://api.tapixai.com/v1/models \
  -H "Authorization: Bearer ***your key"

Returns OpenAI-format data[], each item contains id, owned_by. See full pricing at Pricing page.

Model name quick reference

The usual question is which string to send as model. Below are the most commonly used ones; click to copy. For the full catalog, see Models page.

Case-insensitive (GLM-5.3-Flash and glm-5.3-flash are equivalent); vendor-prefixed syntax is also supported, e.g. 302/qwen3.5-plus.

Error code

Status code	Meaning	How to handle it
400	Invalid parameter	Check the message in the response—most likely the model name or messages format is incorrect
401	Authentication failed	Key is invalid or deactivated—regenerate it in the console
402	Insufficient balance	Retry after top-up
403	No permission	This key is bound to a different model, or the model isn't available to your group
404	Model not found	Call `/v1/models` to confirm the real model name
429	Rate limit triggered	Back off and retry based on the returned Retry-After
5xx	Service error	We automatically switch to a backup channel; if it still fails, we retry with exponential backoff

Rate limits

Default: 60 requests/minute and 200,000 tokens per minute. If that's not enough, you can apply for an increase in the console; for large volumes, go with the enterprise plan.

Relationship with the official API

The API format is identical to the official one, butWe are not official, nor do we represent the official side. Models are provided by each model vendor, prices are set by us, and bills are issued by us. In case of behavioral differences, our documentation prevails; for scenarios requiring absolute official guarantees, we recommend going directly to the official provider.