Model tiers and your own key#

Waper offers Lite, Pro and Max. Choose the tier that fits the task. You can also keep using your own API key or an OpenAI-compatible service: the provider bills you directly, with no Waper credits charged for that model usage.

Choosing a tier#

The button left of send shows the current tier. Open it for three options, each with a short description and a check beside your choice. Models on your own keys and services are at the bottom; until you add one, there is a single Use your own API key item. The web and desktop apps share this picker; on iPhone, open the input box's + menu.

Tier	Best for	Available to
Lite	Quick, economical everyday work	Everyone
Pro	A balance of efficiency and capability for multi-step tasks	Paid members
Max	More thought for complex problems	Paid members

These are capability tiers, not new subscription plans. Both Personal and Pro plans include the Pro and Max tiers. If you haven't chosen, Free starts with Lite and paid members with Pro. Max may use more credits because it spends more on thinking; a task does not have a fixed price per tier.

• Each chat remembers its own choice. Changing it affects that chat.
• It also becomes your default for new groups, new agents' chats and routines. Existing chats keep their own choices.
• Settings → Agent → Models has the same picker for changing your default.
• Platform tiers do not show the underlying model or a thinking-level control. Thinking still appears in the chat and can be expanded afterward.

Free and paid tiers#

Locked Pro and Max options need a paid membership. Selecting one on Free opens the plans window without switching. If your subscription ends, your chosen tier is remembered while the chat uses Lite, as the picker explains. Upgrading again lets you use the original tier. Your own key is not restricted by these tier entitlements.

Switching mid-chat#

You can switch tiers or your own model at any time. Your conversation stays intact.

Your own models and thinking levels#

With your own key or service, the picker still shows the actual model and provider. The composer keeps the maker's icon, or a generic icon for a custom service. Thinking controls remain available for these models: select the adjust button beside the chosen model, or focus its row and press →. Picking a level closes the submenu; ← or Esc returns to the model list.

The options are whichever of Off, Low, Medium, High, XHigh and Max that model supports. With no preference, Low is used, or the nearest supported level, with ties going to the higher one. Switching your own models follows the same rule. Higher levels may take longer and cost more; Max carries a cost notice. This Max thinking level is separate from Waper's Max tier.

Using your own API key#

Three providers are supported: OpenAI, DeepSeek and OpenRouter. Available models are the ones listed at the bottom of the picker.

Adding and checking a key#

1. Open Settings → Agent → Models, find the provider under "Your own API keys" and select Add.
2. Paste the key you created with the provider and select Save.

When you save, Waper first sends one free request to the provider with this key (listing models, or reading the balance or the key's limit). Only a key that passes is saved; if it fails, the reason appears under the box (the key is invalid or revoked, has no credit left, or the provider couldn't be reached) and nothing is stored. So that this can't be used to try out keys in bulk, checks are limited to 10 an hour.

Once saved, only the last four characters are shown ("Connected with key ····abcd"); the full key is never shown again or sent back to your browser. To use a different key, choose Replace key in the row's "…" menu.

Using it#

Once connected, the bottom of the input box's model list shows the models your key can run, each with a "Your key" tag. Pick one and this chat's model calls go through your key from then on. Which models your key can run is whatever that list shows.

With your own key:

• The provider bills you for model usage; no credits are used, so you can keep chatting even with 0 credits. Waper counts this part as 0, and the row in your credit history carries a "Your key" tag.
• Usage that isn't a model chat still uses credits: transcribing audio and video, deep research and summaries of saved links are priced in credits or run on Waper's models, as usual.
• The tidying Waper does for this chat in the background (summaries when it gets long, daily sections) also runs on your key; routines and the weekly memory tidy use your default, which goes through your key when your default is your key.

When the key stops working#

When the provider refuses the key (invalid or revoked), the key runs out of credit or is rate limited, the provider refuses that one request (for example it flags the input, or your region or account can't use the model), or the provider can't be reached, the turn stops and a line in the chat says which provider's key had what problem. Waper does not quietly switch to its own models, and no credits are used. Replace the key in Settings, or switch this chat back to a Waper tier, then send again. A key the provider calls invalid is marked "Not working" in Settings; a key that only had one request refused keeps working.

Deleting a key#

Choose Delete in the row's "…" menu and confirm once. The key is deleted from Waper at once; chats and defaults that used it return to an available Waper tier, clearing the old thinking preference, and the next message works as usual. You can add it back anytime. Deleting your account deletes the key too.

Data and privacy#

• Your key is stored encrypted and used only to call the provider you chose, on your behalf.
• Model requests go from Waper's servers through Cloudflare AI Gateway to the provider with your key, not straight from your browser; requests on your key are not logged by the gateway.
• With your key, what the chat sends to the model is handled and billed by that provider under its own terms. See the Privacy Policy.
• Calls to Waper with an API key or over MCP can't read or change your model keys: those endpoints only take you signed in (see REST API).

Other services (OpenAI-compatible)#

Besides the three providers above, you can add any service with an OpenAI-compatible API: a model you host yourself, a gateway at your company or another model platform, as long as it has a /chat/completions endpoint.

Adding a service#

1. Open Settings → Agent → Models and select Add on "Other services (OpenAI-compatible)", the last row of the "Your own API keys" list.
2. Enter the Base URL (the part before /chat/completions, such as https://api.example.com/v1). Only https:// addresses on the public internet work (on the default port 443, or 8443 or 8080), without a query or password.
3. Enter an API key if the service needs one; otherwise leave it empty. The name defaults to the address's domain.
4. Add the models to use. For each, enter the model ID (the name the service expects), a display name (defaults to the model ID), the context length (128000 by default, at least 50000), the max output (8192 by default, at most 32768) and the thinking levels it takes (sent to the service as reasoning_effort; choose only Off for a model that doesn't think). In a chat, each request may write up to the smaller of the max output and 10% of the context length (at least 8192), thinking included.
5. Select Save.

Models must support tool calling (function calling): the agent reads and writes your knowledge base, searches the web and uses apps through tools. Images and files are read into text by Waper first, so the model doesn't need to see images.

When you save, Waper checks the service first: if it gives a model list, every model ID you entered must be on it; then the first newly added model gets one chat request for a single token with one tool, which proves the key works and the service accepts tool calls (many services hand out their model list without a key, so the list alone proves nothing about the key). The service charges for that request, usually less than a cent. Only a service that passes is saved; otherwise the reason appears under the form (the key is wrong, there's no credit left, the service couldn't be reached or didn't answer like an OpenAI-compatible API, it doesn't have one of the models, or it refused a request with a tool) and nothing is stored. These checks share the limit of 10 an hour with the keys above.

You can add up to 10 services with up to 20 models each. Use the row's "…" menu to Edit or Delete one later; when editing, leave the key empty to keep it, or remove it. Changing the address or key, or adding a model, checks the service again (when you only add models, the request goes to the first new one).

Using it#

Once added, the models on your services are listed at the bottom of the input box's model list after the providers' models, each with the service's name in small text. Pick one and this chat's model calls go to that service from then on.

• The service bills you; no credits are used. Waper counts this part as 0, and the row in your credit history carries the "Your key" tag too. Usage that isn't a model chat still uses credits.
• When the service refuses the key, runs out of credit, refuses that one request or can't be reached, the turn stops and a line in the chat says what went wrong; Waper does not quietly switch to its own models. A service that calls the key wrong is marked "Not working" in Settings; edit it to check it again.
• When you delete a service, or remove a model while editing, chats and defaults that used it return to an available Waper tier, clearing the old thinking preference.

Data and privacy#

• Your key is stored encrypted and used only to call the service you added.
• Requests go from Waper's servers straight to the address you entered, not through Cloudflare AI Gateway, carrying only your key (nothing if you left it empty). Redirects are not followed, and only public addresses are reached.
• What the chat sends to the model is handled by that service under its own terms.

Where your credits go#

The Usage section of Settings → Plan & billing lists your credit history by day: a chat's spending on one day is one row (select it to open that chat; turns on your own key get a row of their own); background tidying, audio and video, deep research and saved-link summaries each get a row, and credits you received are listed too. Spending is negative and credits received are positive; rows that used your own key carry a "Your key" tag. Scroll down to load earlier days. For how credits work and what each plan includes, see Plans.