Documentation / AI and automation
AI
Turn on assistant, analysis, dictation and smart search with a CloseYourIt AI key or your own provider.
CloseYourIt works without AI. The AI features — assistant, ticket analysis and drafts, dictation, smart search and result reranking — turn on when an AI service is connected. Until then they say so, instead of failing.
Open Administration → AI settings and pick one of three sources.
CloseYourIt AI
Paste the key and press Test connection. Addresses and models are ready, reranking included. Requests run on our servers; only how much each key uses is stored, never the text.
While the service is in preview, keys are given by hand with a monthly limit: write to [email protected].
My provider
Any service that speaks the OpenAI API, including a model you host (vLLM, Ollama, LiteLLM).
| Field | Example |
|---|---|
| Address | https://api.openai.com/v1 |
| Key | your provider key |
| Model for text and answers | gpt-4o-mini |
| Model for smart search | text-embedding-3-small (empty = smart search off) |
| Vector size | 1536 — Test connection tells you the real number |
| Model for dictation | whisper-1 (empty = dictation off) |
Reranking service
Optional. After a search, it puts the results back in order from most to least useful. Without it, search still works but the first results are less precise. It uses the /rerank format of Cohere: Cohere, Jina, Voyage and LiteLLM work.
Changing the search model
Each search model writes its own kind of vectors. When you pick a different one, CloseYourIt warns you, then deletes the stored vectors and recalculates them for tickets, errors, ideas and pages. While that runs, smart search finds fewer results. The size cannot go above 2000.
Environment variables
Without a choice on the page, CloseYourIt reads AI_BASE_URL, AI_API_KEY, EMBED_BASE_URL, CHAT_BASE_URL and EMBED_RERANK_MODEL from the server environment.