Connect an OpenAI-compatible API
Many AI services and servers speak the same API format as OpenAI. atbridge.ai works with all of them through one provider option, OpenAI-compatible (custom URL). You don’t install anything or wait for an integration: enter the service’s base URL and API key, and the atbridge AI Chat uses it like any other provider.
Typical cases:
- A model on your own machine or server: LM Studio, vLLM, the llama.cpp server, LocalAI.
- A proxy your company runs, e.g. LiteLLM in front of several providers.
- An API gateway or reseller that gives you access to models through an OpenAI-compatible endpoint.
What you need
Section titled “What you need”- Base URL: the API address from your provider’s documentation. It is the
same value the OpenAI SDK calls
base_url, for examplehttps://llm.example.com/v1. Paste it exactly as documented; atbridge.ai accepts it with or without the trailing/v1. - API key: created in your provider’s dashboard. A local server that
doesn’t check keys accepts any placeholder, such as
local.
The service must support two standard OpenAI endpoints with
Authorization: Bearer authentication: /v1/chat/completions (the chat itself)
and /v1/models (the model list, which atbridge.ai also uses to check the
connection). Almost every OpenAI-compatible service and local server provides
both.
Common local servers and their usual base URLs (check your own setup):
| Server | Usual base URL |
|---|---|
| LM Studio | http://localhost:1234/v1 |
| vLLM | http://localhost:8000/v1 |
| llama.cpp server | http://localhost:8080/v1 |
| LiteLLM proxy | http://localhost:4000 |
Add it to atbridge.ai
Section titled “Add it to atbridge.ai”-
In Thunderbird, open the atbridge Settings: the gear, or Tools → Add-ons and Themes → atbridge.ai → Options (see Open the extension settings). Go to the AI providers tab.
-
In the Provider dropdown, pick OpenAI-compatible (custom URL).
-
Fill in Base URL and paste your key into API key. The key is encrypted on this device and never reaches atbridge.ai’s servers.
-
Click Connect. atbridge.ai checks the address and the key right away by asking for the model list. When the check passes, the status turns to a green ✓; if it fails, nothing is saved and you see the reason.
Pick a model and chat
Section titled “Pick a model and chat”Open the atbridge AI Chat (the @b button in Thunderbird) and pick a model from the model dropdown, which shows the models the service reports. A gateway can list hundreds of models; to use one by its exact name, type it into custom model id under Settings → Preferred AI provider & model.
Who sees your data
Section titled “Who sees your data”With an OpenAI-compatible provider, your prompts, and any email text or attachments you give the AI Chat, are sent to the base URL you entered. atbridge.ai doesn’t proxy or store them. Before you connect a service, check that you trust whoever runs it, and read its privacy and data-retention terms. For mail that must not leave your computer, use a local server (the table above) or Ollama.
Set it up from the command line
Section titled “Set it up from the command line”The same connection from a terminal; the key is read from standard input, so it stays out of your shell history:
echo "$MY_API_KEY" | atbridge connect activate --provider openai-compatible \ --base-url https://llm.example.com/v1 --key-stdinIf it doesn’t connect
Section titled “If it doesn’t connect”- 404 or “not found”: the base URL is wrong. Copy it again from the
provider’s documentation; usually it’s the address that ends in
/v1. - 401 or 403: the key is wrong, expired, or has no access to that model.
- The URL is right but Connect still fails: the service may not provide the
model list (
/v1/models), which atbridge.ai needs to check the connection. Ask your provider whether it supports it. - The chat answers but never looks into your mail: the model doesn’t support tool calling. Pick another model.
Related
Section titled “Related”Was this page helpful?
Thanks for the feedback!