Skip to main content
Presets let you bundle a model, system prompt, and generation parameters into a single reusable configuration. Once created, you reference a preset exactly like a model ID — no extra code, no repeated config. You can also attach a preset to an API key, so its settings apply to every request made with that key. This is the way to control parameters like prompt caching or reasoning effort from apps where you can’t change the request body or the model name. Presets can be private (only your API key can use them) or public (shareable with anyone).

Creating a Preset

Go to Dashboard → Presets → Create Preset and fill in the fields:
1

Name your preset

Give it a short, descriptive name. This is for your reference only.
2

Choose a model

Select any supported model, or choose Any Model to leave the model open at request time. With Any Model, callers must specify the model using the @ syntax — see Overriding the model below.
3

Write a system prompt

Enter the system prompt that should apply to every request using this preset. This is the main reason to use presets — ship a locked, versioned persona or instruction set without exposing the prompt in client code.
4

Set parameters (optional)

Configure defaults like temperature, max_tokens, top_p, reasoning_effort, service_tier, and prompt caching. These apply automatically to every request unless overridden. See Preset settings for the full list.
5

Lock the settings (optional)

By default, a value sent in the request wins over the preset’s. Turn on Locked to make the preset’s values win instead.
6

Set visibility

Choose Private (default) or Public. Public presets can be used by anyone with the preset ID.

Using a Preset

After creating a preset, copy its ID from the dashboard. It looks like:
Paste it into the model field of any Chat Completions request — everything else stays the same.
The preset’s system prompt and parameters are applied server-side — your request payload only needs model and messages.

Overriding the Model

If a preset was created with Any Model, the model must be specified at request time using @ after the preset ID:
This is useful when you want a shared system prompt and parameter set but need the caller to decide which underlying model to use — for example, letting users pick between gpt-4o and gpt-4o-mini while keeping the same persona.

Attaching a Preset to an API Key

Some apps — chat front-ends, roleplay platforms, IDE plugins — only let you enter an API key and pick a model. You can’t add parameters to the request or a suffix to the model name. Attach a preset to the key instead, and its settings apply to every request made with that key.
1

Create a preset

Create a preset as described above with the settings you want, for example prompt caching set to Off.
2

Attach it to a key

Go to Dashboard → API Keys, create a new key or click Edit on an existing one, and select the preset.
3

Use the key as usual

Keep calling regular model IDs like claude-opus-4-6. The app needs no changes.
A key preset works on /v1/chat/completions, /v1/responses, and /v1/messages. The preset’s model sets which requests it covers. A preset with Any Model applies to every request on the key. A preset with a specific model only applies to requests for that model, and requests for other models are left untouched. Settings a model doesn’t support are skipped, not rejected. For example, reasoning_effort is ignored on models without reasoning, and a service_tier the model doesn’t offer is ignored. This lets one Any Model preset serve every model on the key without causing errors.
The system prompt is only applied when you call the preset as a model (presets/<id>). A preset attached to a key applies its parameters, not its system prompt.
On /v1/messages, max_tokens is required in every request, so a key preset’s max_tokens only replaces it when the preset is Locked.

Preset Settings

Prompt caching

For models where Routeway places cache breakpoints for you (Claude models), Routeway caches with a 1-hour lifetime by default. prompt_cache changes that:
  • off — no cache breakpoints are sent, including any your app adds itself. You pay no cache-write cost, but you also get no cheap cache reads. Choose this when consecutive requests rarely share the same opening, for example when you often edit earlier messages, regenerate, or reply with long pauses.
  • 5m — 5-minute cache. Cache writes cost less than with 1h. Good when you send the next request within a few minutes.
  • 1h — 1-hour cache, the default.
Models that cache automatically on the provider’s side, such as DeepSeek, aren’t affected by this setting. See Prompt Caching for pricing details.

Streaming with Presets

Presets work with all standard Chat Completions parameters, including stream:

Public vs Private Presets

Public presets expose the system prompt to anyone who has the preset ID. Don’t include secrets, internal instructions, or sensitive information in the system prompt of a public preset.

When to Use Presets

Ship locked personas

Define a customer support agent, coding assistant, or brand voice once and reference it by ID across all your apps.

Version your prompts

Create a new preset when you iterate on a system prompt instead of updating in-place, so you can roll back.

Share with teammates

Public presets let anyone on your team (or your users) call a consistent configuration without needing to manage the prompt themselves.

Keep clients simple

Offload model selection and prompt engineering to the dashboard. Client code only needs the preset ID.