Skip to main content
KeyRing Docs
Documentation/Model Config

Model Config

Use the Model Config module to inspect defaults, apply presets, edit per-model overrides, and save advanced runtime settings.

Read time: 11 min read

What Model Config Does

Model Config is KeyRing's advanced per-model override editor. It lets you select a provider and model, load recommended defaults, layer on overrides, preview the merged effective configuration, and save those overrides for future requests.

Power-user module

If API Settings decides which model a provider uses, Model Config decides how that model behaves once it has been selected.
KeyRing AI Model Config module
Model Config is the advanced override editor for provider-model behavior, presets, and merged runtime settings.

Layout

The current layout is a three-pane workspace with draggable dividers. Pane widths are remembered locally by the app.

Left pane

Provider filter, provider list, model filter, and model list.

Center pane

Profiles, one-click presets, quick toggles, hints, and Live JSON.

Right pane

The detailed tabbed editor for decoding, safety, retrieval, performance, and transport.

Top Action Bar

  • Reset to Recommended
  • Import JSON
  • Export JSON
  • Save Changes
  • Apply to All
  • Close (Esc)

Keyboard shortcuts currently include Ctrl+S, Cmd+S, Ctrl+Enter, and Cmd+Enter for saving the current override.

Select a Provider and Model First

1

Choose a provider in the left pane

Select the provider you want to tune.

2

Choose a model from that provider

Select the exact model, not just the provider family.

3

Let the editor load defaults and saved overrides

The editor becomes useful only after both the provider and model are selected.

Profiles, Presets, and Quick Toggles

The center pane gives you fast-start tools before you move into the detailed tab set.

Profiles

Profiles are backend-fed reusable starting points. They may vary by build and are intended as a baseline, not always a final configuration.

Built-in one-click presets

  • SCIENCE MODE pushes the config toward lower-temperature, more controlled output.
  • CREATIVE MODE increases generative freedom and more expressive decoding.
  • MAXIMUM I/O raises input/output ceilings and extends timeout aggressively.

Quick Toggles

  • Stream
  • Cache
  • Use Web/Search
  • Use RAG
  • Safety Filters

The Detailed Tabs

The right pane exposes the detailed override controls. Each tab groups a different class of model behavior.

Overview: system prompt, input/output token caps, timeout, response format, grammar, and JSON schema.

Decoding: strategy, temperature, top-p, top-k, min-p, typical-p, TFS, Mirostat, penalties, and temperature annealing.

Beam / N-best: beams, early stopping, best-of, n, and diversity penalty.

Output Control: stop sequences and max-sentence limits.

Tools & Routing: tool choice, parallel tool calls, tools JSON, and routing heuristics.

Safety: safety enablement, prompt injection blocking, self-reflection leak control, and shell-command disallow rules.

Retrieval / RAG: web/search toggle, RAG toggle, top-k values, reranking, relevance thresholds, and citations.

Performance: stream, cache, speculative decoding, batching, truncation, and prefill cache.

Logging / Transport: logprobs, seed, logit bias, extra headers, and metadata.

Hints and Live JSON

The center pane is one of the most useful parts of the module because it helps you see the effective configuration rather than just the form fields.

  • Hints shows provider-aware notes and practical warnings.
  • Live JSON (Merged, effective config) shows the merged output after KeyRing combines backend defaults, saved override, and current unsaved edits.

Save, Reset, Import, Export, and Apply to All

Save Changes stores the override for the selected provider/model pair only.Apply to All sends the current override payload across all models through the backend bulk-apply path.

Use Apply to All carefully

Apply to All is much broader than a normal save. Confirm the current working payload is intended to be global before using it.

Recommended Workflow

  1. Select the provider and exact model first.
  2. Review the current defaults before changing anything.
  3. Apply a profile or preset only if it matches your goal.
  4. Use the detailed tabs for precise adjustments.
  5. Watch the Live JSON panel while you work.
  6. Save to the single provider/model first.
  7. Only use Apply to All when you are certain the change should be global.

Troubleshooting

Common issues

  • The editor looks empty: Make sure both a provider and a model are selected.
  • JSON behavior feels forced: When a JSON schema is present, JSON-oriented guardrails are expected.
  • A provider ignores one of your settings: Provider support still varies; use the Hints panel and real request testing as the final source of truth.