AI provider settings

Ollama (local) active

Host Model Pick from what's pulled on this host… deepseek-r1:14bdeepseek-r1:32bgemma3:27bgemma4:12bllama3.1:8bllama3.3:70bllava:13bmistral-small3.1:24bnomic-embed-text:latestphi4-reasoning:14bphi4:14bqwen2.5-coder:14bqwen3:14bqwen3:30b-a3bqwen3:32b Save

Test connection

Set active

Custom analysis instructions

Optional free-text guidance applied on top of the built-in extraction/matching prompts, regardless of which provider above is active — e.g. "treat any call mentioning maritime logistics as especially relevant" or "ignore calls under €50k per project."

Save

Claude (Anthropic API)

API key Model Save

Test connection

Set active

OpenAI

API key Model Save

Test connection

Set active

Local model benchmark (Ollama)

A real extraction call against one fixed sample EU call, run once per locally pulled model — lets you pick a model that's actually fast AND reliably tags domains/technologies correctly on this machine, rather than guessing from size alone. A model's very first call after being switched to includes a one-time cost loading it into VRAM — if a result looks surprisingly slow, re-run once more before ruling it out.

Re-run benchmark