Custom APIs & Local LLMs
Set up the Custom summary provider with a compatible local, network, or internet-hosted chat-completions endpoint.
Where transcript text goes
Meetily sends transcript text to the endpoint you configure. A localhost endpoint keeps processing on this device; network and internet endpoints send it to that server, so that server's access, retention, and privacy rules apply.
Meetily can use a custom endpoint for AI summaries when the endpoint accepts its OpenAI Chat Completions request and authentication shape, and returns a compatible response.
Use this guide after selecting Custom in Summary providers. For other summary options, including Built-in AI, hosted providers, and Ollama, return to that page.
Open Custom API settings
You can reach the same summary-provider settings from any of these places:
| Starting point | Path |
|---|---|
| Meeting summary actions | Select the AI Provider settings gear. |
| Meeting Preferences | Preferences > Summary Model |
| App settings | Settings > Summary > AI Provider |
How the endpoint is used
Meetily treats the URL you enter as a base endpoint and appends /chat/completions. The server therefore needs to support the OpenAI Chat Completions request and response format that Meetily uses.
Many compatible servers use a base URL ending in /v1, for example:
http://localhost:1234/v1That becomes:
http://localhost:1234/v1/chat/completionsThe required base path is set by the server, not by Meetily. For example, Open WebUI exposes its compatible chat-completions route below /api, so its base URL is different. Check the server's own documentation before saving a URL.
Configuration fields
| Field | Required | Description |
|---|---|---|
| Endpoint | Yes | The server's base URL, including the version or API path it requires. |
| Model | Yes | The exact model ID the server expects. |
| API key | When required | The key used by servers that require authentication. |
| Max tokens | No | The maximum response length. |
| Temperature | No | Controls how varied the generated summary can be. |
| Top-p | No | Controls the sampling range used for generation. |
Configure a custom endpoint
Start the server you want to use.
Find its base endpoint and exact model ID in the server's interface, startup output, or documentation.
Open Custom API settings from one of the three paths above.
Enter the endpoint, model, and API key when the server requires one. Adjust max tokens, temperature, and top-p only if you need to override the server defaults.
Save the settings, then generate a test summary from a non-sensitive meeting to confirm the connection and response shape.
Server guides
LM Studio
Run a local OpenAI-compatible server from LM Studio's Developer tab.
Ollama
Use Meetily's Ollama provider or Ollama's OpenAI-compatible API.
Open WebUI
Connect to Open WebUI's API endpoint with an API key.
vLLM
Serve a model with vLLM's OpenAI-compatible server.
LM Studio
LM Studio can run a local server from its Developer tab. Download or load a model, start the server, then use the model identifier shown by LM Studio.
Its OpenAI-compatible API normally uses this base URL on the default port:
http://localhost:1234/v1LM Studio also has a native /api/v1 API. Use its OpenAI-compatible /v1 base URL here because Meetily sends Chat Completions requests.
Ollama
Ollama has an OpenAI-compatible /v1/chat/completions API. Its normal local base URL is:
http://localhost:11434/v1Meetily also provides an Ollama summary provider. Use that provider for the direct Ollama integration; use a custom endpoint only when your setup specifically needs the OpenAI-compatible route. Ollama's compatibility examples use an API key value that is required by some clients but ignored by the local server.
Open WebUI
Open WebUI exposes an OpenAI-compatible chat-completions route at /api/chat/completions. With the usual local port, enter this base URL in Meetily:
http://localhost:3000/apiCreate an API key in Open WebUI and enter it in Meetily. Select the exact model ID that Open WebUI makes available to that key.
vLLM
vLLM provides an OpenAI-compatible server. Start one with a model, for example:
vllm serve meta-llama/Llama-3.1-8B-Instruct --port 8000Use the base URL and model name that the server reports. With the command above, the usual base URL is:
http://localhost:8000/v1Other compatible servers
Jan
Jan's Local API Server runs an OpenAI-compatible API. Start it from Settings > Local API Server, set its required API key, and use its configured prefix. The default base URL is http://127.0.0.1:1337/v1.
LocalAI
LocalAI exposes an OpenAI-compatible API. After starting LocalAI and choosing a model, its standard chat-completions route is under /v1; the typical base URL is http://localhost:8080/v1.
text-generation-webui
text-generation-webui supports OpenAI-compatible chat and completions APIs. Enable its API when starting the server, then use the /v1 base path and port shown by its startup configuration. The traditional default is http://127.0.0.1:5000/v1.
llama.cpp server
llama.cpp's llama-server exposes an OpenAI-compatible Chat Completions API. Start it with a chat-capable model, then use its configured port and /v1 base path. On the default port, that is http://localhost:8080/v1.
Troubleshooting
Check the server's documented Chat Completions path. Meetily appends /chat/completions, so enter the base path that produces the server's full route. A /v1 base is common, but Open WebUI is an example that uses /api instead.
Confirm the server is running, the hostname and port are reachable from this computer, and firewall or network rules allow the connection. Test with a non-sensitive summary after changing the URL.
Copy the model ID exactly as the server reports it. A downloaded model name, a file name, and the server's exposed model ID are not always the same.
Enter the server's API key if it requires one. Local servers vary: Jan requires a key for its local API server, while Ollama's local OpenAI-compatible endpoint ignores the compatibility key used by some clients.
Verify that the server supports OpenAI Chat Completions requests and returns a compatible chat-completion response. Check the server logs for a rejected request, unsupported model, or unsupported generation parameter.
Last updated on