Configuration

★

Presets and saved statesStart from a ready-made look, or save and reload your own setups

A preset changes colors, fonts, background and shape in one click. You can still adjust anything afterwards. Changes save by themselves.

Saved looks

Save the current look under a name

Bring a saved look back

Manage saved looks

Import

Remove or reset…

⚙

ModelWhere the answers come from

Choose the AI that writes answers. Demo works offline with no key. Paste your key below. It is saved in a private file on this computer and never shown again.

API key

ModelPick from the list your provider offers.

Base URL (Ollama and OpenAI-compatible servers only)Where the server runs, for example http://localhost:11434 for Ollama.

Temperature Low is precise and repeatable. High is more creative. Sources per answer How many passages are shown to the model for each question.

Rewrite follow-up questions before searching

↗

Portable appMove this app to another computer

Export app.yaml, the current model settings, saved look, prompts and configured local documents. Keys, accounts, conversations and personal uploads stay here. Review included documents before sharing. Python plugins and remote services still need their original setup.

↻

Document freshnessKeep local evidence up to date

Local folders are checked when someone asks a question, at most once every 30 seconds by default. Changed, added and removed documents update the search index. No background polling or model calls. Set refresh_interval in the local source YAML to change the interval.

⚙

BudgetLimit public chat usage

Daily limits reset at midnight UTC and survive restarts. 0 means no quota. Model-call limits cover answer-pipeline calls, including follow-ups and provider fallback as one invocation. Failed attempts count. Configuration tests/generation and provider-side retries are not counted individually.

Questions per day, whole app Questions per user per day Model calls per day Dollar cap (USD)

Dollar accounting is not available yet. Setting any dollar cap blocks non-Demo answer-pipeline calls rather than guessing a bill. It does not limit owner-run configuration tests, external search fees or embedding costs. Set provider-side billing limits before publishing.

✓

All set?A quick check before people use it

Everything here saves by itself. This is the current setup at a glance.

Open your chat