Configuration

★

Presets and saved setupsChoose an appearance preset

A preset changes appearance only. Appearance edits save automatically. A named setup also stores model settings; loading it replaces the current appearance and model settings. Prompts, sources and keys are not restored.

Saved setups

Save a new setupReusable appearance and model settings. Prompts, sources and keys are not included.

Restore a saved setupRestoring replaces the appearance and model settings included in that setup.

Manage saved setups

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.

Optional check: sends one small request to this provider and model. Your provider may charge. It does not test sources or save settings.

Base URL override (optional)Leave blank for this provider's default. For a custom server, enter its API address.

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

Answer output limitMaximum tokens per model response, not a target length. Higher limits may cost more. Deep searches up to twice the selected passage count. Rewrite follow-up questions before searching

↗

Load and share NerdsLoad saved settings, or download a shareable ZIP

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.

Drop your bundle here

ZIP restores bundled settings, prompts, helper drafts and documents. YAML fills its declared settings; extra files need a ZIP.

Choose a fileZIP or YAML. 1. Choose 2. Review included files and code 3. Run

Share this Nerd

Download an importable bundle of this app and its local documents. Keys and chats stay here.

Include my saved API keys (sensitive)

↻

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. Build with AI, Ask AI and deployment-plan calls count too. Provider-side retries may incur additional usage.

One submitted chat question can use several model calls for searches, the answer, checks and follow-ups. Question limits apply to the whole app or one user as labelled; the model-call limit is shared by the app. 0 means no daily limit. Usage resets at midnight UTC.

No verified provider billing meter is available. Leave Dollar cap blank for no local dollar cap. Any cap, including 0, blocks non-Demo model calls covered by this guardrail; use provider-side billing limits too.

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. When a daily quota is reached, covered requests stop until midnight UTC or the owner changes the limit. Saving takes effect immediately; no restart is needed. Set provider-side billing limits before publishing.

✓

Workspace setupOne setup status page. Choose a group and a tab to edit one panel at a time.

Open your chat