Aimogen ProDocumentation
Product page Support

Custom OpenAI-compatible APIs

Pointing Aimogen Pro at any OpenAI-compatible endpoint, adding models manually, and what compatibility actually requires.

Aimogen Pro can send its OpenAI-format requests to any endpoint that speaks the same wire protocol. This covers self-hosted runtimes such as vLLM, LM Studio, LocalAI, text-generation-webui, corporate gateways, and providers that offer an OpenAI-compatible shim.

Configure#

Aimogen Pro › Settings › API Keys

SettingValue
Main API Service Provider Selectorcustom
Custom OpenAI Compatible API Base URLThe base URL, for example https://llm.internal.example.com
OpenAI / AiomaticAPI API KeysThe bearer token the endpoint expects. If it needs none, enter any placeholder
Additional AI Model List (One Per Line)The models your endpoint serves

The base URL should be the root the API is served from. The plugin appends the standard OpenAI paths, so a base of https://llm.internal.example.com produces https://llm.internal.example.com/v1/chat/completions.

Aimogen Pro > Settings > API Keys with Main API Service Provider Selector set to custom

Adding models manually#

The Additional AI Model List (One Per Line) field exists because the plugin cannot know what a custom endpoint serves. Each line becomes a selectable model:

my-org/llama-3.3-70b
my-org/qwen2.5-32b
mixtral-8x7b

The list is stored in the aiomatic_custom_models option and merged into every model selector.

What "compatible" has to mean#

At minimum the endpoint must implement:

POST {base}/v1/chat/completions

accepting model, messages, max_tokens, temperature, top_p and returning the standard choices[].message.content shape.

For the full plugin feature set it should also support:

PathNeeded for
POST /v1/embeddingsEmbeddings and RAG
POST /v1/images/generationsImage generation through this slot
Streaming (stream: true, SSE)Chatbot streaming responses
tools / tool_callsChatbot Extensions and agents

Endpoints that implement only chat completions still work for content generation; features whose prerequisites are missing will fail with errors from your endpoint rather than from the plugin.

Typical setups#

vLLM

bash
python -m vllm.entrypoints.openai.api_server \
  --model meta-llama/Llama-3.3-70B-Instruct \
  --port 8000

Base URL http://your-server:8000, model meta-llama/Llama-3.3-70B-Instruct.

LM Studio — start the local server, base URL http://127.0.0.1:1234, and add the model identifier LM Studio shows.

A corporate gateway — base URL as supplied by your platform team, and the models they expose.

Reliability behaviour#

Models routed through the custom slot are classified as provider custom. They participate in Reliability Mode like any other provider: they can be cooled down after repeated failures and can be failed over to and from.

Security#

  • Use HTTPS. The bearer token and every prompt travel over this connection.
  • If the endpoint is on a private network, keep it there and restrict it to the WordPress server.
  • Requests carry the same content as any other provider request — see Data, GDPR and retention.

Common problems#

404 on every request Base URL includes a path it should not, or the endpoint does not serve /v1/chat/completions. Test with curl.

A model is not in the dropdown Add it to Additional AI Model List.

Works in curl, fails from WordPress Outbound restrictions, a self-signed certificate, or a firewall between the servers. Check System & Logs › Activity Logs for the transport error.

Streaming does not work The endpoint does not implement SSE streaming. Disable streaming for the chatbot.

Still stuck? Open a support ticket and include the diagnostics from Aimogen Pro › System & Logs › System Info.