AI providers
Which model jlog uses, what it is used for, and how your key is kept.
Bring your own model
jlog does not resell model access. You choose a provider in Settings → LLM provider and use your own key, billed by that provider at their price. jlog never marks it up. A whole job search usually costs under €2 in model usage.
Supported providers
| Provider | Default model | Needs a key |
|---|---|---|
| Anthropic | claude-3-5-haiku-20241022 | Yes |
| OpenAI | gpt-4o-mini | Yes |
| Gemini | gemini-1.5-flash | Yes |
| Ollama | llama3 | No, it runs on your own machine |
You can change the model name for any of them. For Ollama, set the server’s URL (by default http://localhost:11434) and pick from the models it reports.
What it is used for
- Extract with AI in the extension, to read a posting. See capturing applications.
- CV import, to structure the text of a PDF. Without a provider, jlog’s own rules do it.
- Tailored CVs and drafted answers, on Pro.
Tracking, capture on the six boards, and autofill never call a model.
How your key is kept
Your key is encrypted with AES-GCM before it is stored, and is sent nowhere except the provider you picked. The settings page shows only that a key is saved, never the key itself. Save a new one to replace it.
Ollama and the hosted version
The model is called from jlog’s API, not from your browser. On the hosted version the API runs on Cloudflare, so http://localhost means Cloudflare’s machine, not yours: your Ollama server has to be reachable from the internet, for example through a Cloudflare Tunnel. When you run jlog locally, localhost works as expected.