Which AI provider should I pick?

MetaFrazo's AI features — Deep Analysis on individual visuals, the support assistant — can run on three different backends. The choice is yours, per organization, and you can change it at any time. The right pick depends on how strictly your organization controls data egress and how much you want the latest frontier-model quality.

The three options at a glance

• MetaFrazo AI (the default). Your queries and the data they reference are analyzed by an EU-hosted open-weight model, run for you by MetaFrazo under a contractual zero-data-retention and no-training term: your inputs and outputs are not stored after the request and are never used to train models. The data is processed in the EU. You don't supply or pay for any AI key — MetaFrazo operates this backend on your behalf.
• OpenAI. You bring your own OpenAI API key. Queries and the data they reference are sent to OpenAI for analysis. OpenAI's data-handling terms apply.
• Anthropic. You bring your own Anthropic API key. Same trade-off as OpenAI — queries and the relevant data go to Anthropic, with Anthropic's data-handling terms.

You can also leave every provider deactivated — in that mode, MetaFrazo's AI features are simply unavailable for your organization, and the dashboard surfaces a clear "AI is disabled for this organization" message instead of running any inference. Some compliance contexts require this.

When MetaFrazo AI is the right pick

MetaFrazo AI is the default for a reason. It is the right pick when:

• Your organization requires that AI processing stay in the EU under a no-retention term. MetaFrazo AI's processor is a single EU AI provider, disclosed in our sub-processor list, operating under a contractual zero-retention and no-training term.
• You operate under GDPR, NIS2, or DORA and need a defensible "where does our data go" answer for your auditors. With MetaFrazo AI, the answer is: processed in the EU, on independently ISO 27001-certified European infrastructure, under a contractual zero-retention and no-training term. (The EU location reflects how our provider operates the service; the zero-retention and no-training commitments are contractual.)
• You don't need the absolute latest frontier-model quality. MetaFrazo AI runs on an open-weight model that produces solid, useful summaries for Jira-event analysis — but if you want the bleeding edge of language-model capability, the commercial vendors are still ahead.

When OpenAI or Anthropic might be the right pick

Switching to OpenAI or Anthropic is right when:

• Your team is already comfortable sending operational data to that vendor and you have an existing data-processing agreement in place.
• You want maximum analysis quality on each Deep Analysis run and are willing to accept the additional data flow.
• You're already paying for OpenAI / Anthropic capacity and want to use it across your tooling consistently.

Either way, you supply the API key yourself. MetaFrazo never holds a commercial-AI API key on your behalf; the key is stored encrypted in our European backend and is used only to authenticate your organization's calls to the vendor.

Controlling cost on OpenAI / Anthropic

Because OpenAI and Anthropic bill you for every analysis on your own key, a single Deep Analysis on a data-heavy visual can send a large amount of chart data to the vendor — which costs more on an expensive model. To keep that predictable, each of these providers has a Max input tokens setting in its Configure dialog on the Integrations page.

• It caps how much of a visual's underlying data is sent to the vendor per analysis. If a visual holds more than the cap, MetaFrazo analyzes the most relevant portion and tells the model that only part of the data was included, so summaries stay honest rather than silently dropping rows.
• Leave it blank to use a sensible default. Set it lower for an expensive model to tighten the per-analysis cost, or higher if you want the model to see more data and don't mind the spend.
• It's an organization-level setting that only an owner or admin can change, and it applies to that provider only. MetaFrazo AI has no such field — MetaFrazo covers the cost of the default backend, so there's no per-call vendor bill for you to cap.

What "switching providers" actually means

Provider selection happens at the organization level, in the dashboard's Integrations view. Only an owner or admin can change the active provider. The change is immediate — the next Deep Analysis run uses the new provider.

Activating any provider — including MetaFrazo AI — shows an explicit acknowledgement screen explaining where your AI data is processed and under what terms (for MetaFrazo AI: an EU AI service under a contractual zero-retention, no-training term; for OpenAI / Anthropic: that vendor, under that vendor's terms). You have to accept the acknowledgement before the change takes effect. We do this because the change has real compliance consequences and we don't want it to happen by accident.

If you change your mind, switching back to MetaFrazo AI takes effect on the next request.

What about a brief outage?

Operational AI services occasionally have brief outages. MetaFrazo AI has no backup vendor: if the EU AI service is unavailable for a short window, Deep Analysis is briefly unavailable rather than rerouting your data somewhere else. Your data never silently moves to a different backend to cover an outage — the boundary you chose holds even when the service is down. (If you've switched to OpenAI or Anthropic, an outage there behaves the same way — the request fails rather than falling back elsewhere.)

Disabling AI entirely

If your organization's policy is "no AI inference at all", deactivate every provider. The dashboard's AI features will simply show a "disabled for this organization" message, and no inference happens — neither MetaFrazo AI nor any external vendor. You retain access to every non-AI visual without compromise.

Where to go from here

• How is my organization's data isolated from other tenants? — the broader story on data scoping.
• Roles and permissions — confirm who in your organization can change AI settings.
• GDPR, NIS2, DORA — what does MetaFrazo give me? — how AI-provider choice feeds your compliance posture.