Skip to main content

Which AI models does REQQA use, and who pays for them?

The short answer: the AI model is configured per organisation, and each organisation supplies and pays for its own API key. REQQA does not bundle a shared AI account. When you analyse a requirement, generate stories, or run any other AI-driven task, the call is made with your organisation's key, against the model your organisation has chosen.

This page explains what's configurable, where you set it, and how usage is recorded.

Which models are supported​

REQQA can call models from two providers, and it works out which provider to use from the model name itself:

  • Model names beginning with claude- are routed to the Anthropic API.
  • Model names beginning with gpt- are routed to the OpenAI API.

The Organisation Settings page offers a drop-down of known models. At the time of writing it includes several Claude models (Sonnet, Opus and Haiku variants) and several OpenAI models (the GPT-4 family). There is also an Other option, which lets you type an exact model identifier by hand if the model you want isn't in the list yet — useful when a provider releases a new model before the drop-down is updated.

note

The drop-down is a convenience, not a hard limit. Because the provider is detected from the name prefix, any valid claude-* or gpt-* identifier your key has access to can be entered via Other. If a model name doesn't start with a recognised prefix, REQQA can't tell which provider it belongs to and the call will fail.

You don't have to think about "OpenAI versus Anthropic" as a separate setting — you choose a model, and the routing follows.

Who pays — your key, your account​

REQQA is multi-tenant, and AI billing is per organisation:

  • Each organisation stores its own API key (OpenAI or Anthropic) in its settings.
  • Every AI call REQQA makes on your behalf uses that key.
  • The cost of those calls is billed by the provider directly to the account that owns the key — not to REQQA.

In other words, you bring your own key. Without a key configured, an organisation's AI features simply won't run: there is nothing for REQQA to authenticate with. (Some organisations in a shared environment may have no key set, in which case their AI calls fail until one is added.)

tip

If you're trialling REQQA and don't yet have an API key, the non-AI parts of the product — writing your mission, capturing requirements, organising releases — all work without one. You only need a key when you want to run analysis or generate content.

What you can configure​

Open Organisation Settings to manage the AI configuration for your organisation. The settings that affect AI behaviour are:

SettingWhat it does
API keyThe OpenAI or Anthropic key used for all of this organisation's AI calls.
AI modelThe model identifier (e.g. a claude-* or gpt-* name). Determines both the model and, via its prefix, the provider.
TemperatureControls how deterministic the output is. REQQA constrains this to a sensible range (0.0–2.0) and defaults low, because requirements analysis benefits from consistent, repeatable output rather than creative variation.
Max tokensAn upper bound on the size of each response, within a permitted range.
Retry settingsHow many times, and with how much delay, REQQA retries a transient API failure before giving up.

When you enter or change the API key, REQQA validates it by making a small test call before saving — so a mistyped or wrong-provider key is caught at the point you set it, not later when an analysis fails.

How your key is protected​

API keys are encrypted before they are stored and decrypted only when a call is about to be made. They are never displayed back to you in full once saved — the settings page shows only whether a key is present, not the key itself. See Your data and privacy for more on how REQQA handles sensitive data.

How usage is recorded​

Every AI call REQQA makes is logged for your organisation. Each log entry records the model used, the prompt and response, the time taken, and the token counts (prompt tokens, completion tokens, and total). Because token usage is exactly what providers bill on, this log is your audit trail: you can see which activities consumed AI, against which model, and at what token cost.

This is recorded per organisation, so usage and cost stay isolated between tenants — consistent with REQQA's wider multi-tenant model, where everything is scoped to an organisation.

note

REQQA records token counts, not currency amounts. It deliberately does not quote or calculate prices, because per-token pricing is set by the AI providers and changes over time. To turn tokens into a monetary figure, consult your provider's current pricing for the model you've selected.

Choosing a model — practical guidance​

There's no single right answer; it's a trade-off your organisation makes:

  • Larger, more capable models tend to produce sharper analysis and more faithful generated stories, at a higher per-token cost.
  • Smaller or faster models cost less per token and respond quicker, which can be fine for lighter tasks or high-volume runs, at some cost to depth.
  • Temperature is best kept low for analysis work — you generally want the same requirement to produce the same findings each time it's analysed.

Because the setting lives at the organisation level, you can change models as your needs and budgets evolve, and the change takes effect for subsequent AI calls without any other reconfiguration.