An LLM profile is the chat model that writes the answers — the "voice" of your assistant. Where the embeddings profile finds the relevant content, the LLM profile turns what was found into a reply. You can keep several profiles (say, a fast inexpensive model for a public chat and a stronger one for an internal tool) and pick per chat which one to use.
The LLM Profiles tab #
Go to Wyregraf → Options and open the LLM Profiles tab.
- The LLM Profiles panel lists your profiles — add, edit, or delete them here.
- Default LLM profile is the profile new chats start with.
Profile settings #
Each profile connects one provider account and model:
- OpenAI — enter your OpenAI API key and pick a model. The list covers the current OpenAI lineup, from the flagship GPT-5.5 down to the mini and nano tiers;
gpt-4.1-miniis the default and a good balance of quality, speed, and cost for answering questions from retrieved content. - Anthropic — enter your Anthropic API key and pick one of the supported Claude models:
Claude Sonnet 4.6 for the best balance of quality, latency, and cost; Claude Haiku 4.5
for the fastest, lowest-cost option; or Claude Opus 4.6 for the most complex work.
The remaining fields (temperature, max tokens, and the other sampling controls) are tuned for retrieval answering out of the box — leave them alone unless you know you want different behavior.
Which model should I pick? #
For a documentation or support assistant, start with the default. Retrieval does the heavy lifting — the model mostly rephrases content your own site provided — so a mid-tier model usually answers as well as a flagship one at a fraction of the cost. Step up a tier if you find answers miss nuance across several retrieved passages.
Provider availability and prices can change. Use Test connection after selecting a model and
before publishing the chat. Changing the chat model does not require rebuilding an embeddings index.
What to do next #
With both profiles saved, create your first embeddings — that is where your content gets indexed.
