OpenAI-compatible endpoint

Point any OpenAI-style client at https://connectorbay.com/v1 and ConnectorBay connectors become tools your model can call. You bring two kinds of key (BYOK): one for the model provider you choose, and one for each connector that needs it. We run the tool calls for you and return the final answer in the standard chat completions format.

Try it

curl https://connectorbay.com/v1/chat/completions \
  -H 'Authorization: Bearer <your model provider key>' \
  -H 'X-Model-Provider: groq' \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "llama-3.3-70b-versatile",
    "messages": [{"role": "user", "content": "Which Groq models can I use?"}],
    "connectorbay": {"connectors": ["byok-groq"], "keys": {"byok-groq": "<your Groq key>"}}
  }'

The same call with the OpenAI Python library:

from openai import OpenAI

client = OpenAI(base_url="https://connectorbay.com/v1", api_key="<your model provider key>",
                default_headers={"X-Model-Provider": "groq"})
r = client.chat.completions.create(
    model="llama-3.3-70b-versatile",
    messages=[{"role": "user", "content": "Which Groq models can I use?"}],
    extra_body={"connectorbay": {"connectors": ["byok-groq"], "keys": {"byok-groq": "<your Groq key>"}}})
print(r.choices[0].message.content)

How it works

Send a normal chat completions request. Add a connectorbay object with the connectors you want (for example byok-groq) and, where a connector needs one, its key. We add those connectors' tools to your request, call the model, run any tool calls it makes against the connectors, give the results back to the model, and repeat until it answers. The response also lists which connector tools ran, under connectorbay.tool_calls. Tool lists for every connector are public at /openai/index.json.

Headers and fields

Authorization: Bearer <key> is your model provider key and is passed to the provider. X-Model-Provider picks the provider: openai, groq, openrouter, together, mistral, deepseek, fireworks, xai, perplexity, cerebras, sambanova. The default is openai. connectorbay.connectors is a list of up to 5 connector names. connectorbay.keys maps a connector name to its key, or to a pair of keys for connectors that need two.

Limits

Responses are not streamed yet, so send stream: false. A request may use up to 5 connectors and 64 tools in total, and the model gets at most 4 rounds. Each request is capped at about a minute and 4,096 output tokens. Tools are read-only. If the model calls a tool of your own that is not one of ours, the call comes straight back to you, as with any OpenAI endpoint. Servers on your own machine are not reachable from here.

Your keys

Keys are used for the one request and are never stored, logged or cached. The model key goes only to the provider you named. A connector key goes only to that connector's provider through our hosted connector. We cannot see the cost of your model calls, which your provider bills to you. This endpoint is new, so please tell us at agent@connectorbay.com if something does not behave like OpenAI's.