Providers & subprocessors

This page is generated from mumo's live routing configuration each time it is rebuilt. It lists every inference provider that receives your content and the models you can select from each, the platform features that send content to a provider you did not pick, and the other service providers that process data for us.

We route only to (a) model developers serving their own models under API terms that prohibit training on submitted content, or (b) inference providers running open-weight models on their own infrastructure. We do not route through gateways, routers, or intermediaries, and we never send requests to the developer of an open-weight model.

For providers serving their own models, the no-training commitment is a term of the provider's standard API agreement, which we accepted when we opened our account — binding on both of us, but their standard terms, not a negotiated contract; marked "API terms" below and linked to the source. For inference providers running open-weight models, we rely on the provider's published data-handling statements — marked "Stated policy" below and linked to the source.

1. Inference providers

Every provider that receives the content of a conversation you start, with the models you can currently select from it. A model we no longer offer can still run in a session that started on it; it stays with the same provider and is counted here, not named.

ProviderServesModelsReceivesNo-training commitmentLast verified
AnthropicIts own modelsHaiku 4.5, Opus 5, Sonnet 5 (+6 no longer offered)prompt; selected context (shared peer answers, forwarded snippets, web sources); generation parameters; technical metadataAPI terms2026-08-03
GoogleIts own modelsGemini 3.1 Flash Lite, Gemini 3.7 Flash (+7 no longer offered)prompt; selected context (shared peer answers, forwarded snippets, web sources); generation parameters; technical metadataAPI terms2026-08-27
MetaIts own modelsMuse Spark 1.2 (+1 no longer offered)prompt; selected context (shared peer answers, forwarded snippets, web sources); generation parameters; technical metadataAPI terms2026-08-27
OpenAIIts own modelsGPT-5.6 Luna, GPT-5.6 Sol, GPT-5.6 Terra (+10 no longer offered)prompt; selected context (shared peer answers, forwarded snippets, web sources); generation parameters; technical metadataAPI terms2026-08-03
xAIIts own modelsGrok 4.6 (+6 no longer offered)prompt; selected context (shared peer answers, forwarded snippets, web sources); generation parameters; technical metadataAPI terms2026-08-03
FireworksOpen-weight models, run on its own infrastructureDeepSeek V4 Flash, DeepSeek V4 Pro, GLM 5.2, Kimi K3, MiniMax M3, Qwen 3.8 Max (+7 no longer offered)prompt; selected context (shared peer answers, forwarded snippets, web sources); generation parameters; technical metadataStated policy2026-08-27
Together AIOpen-weight models, run on its own infrastructureGLM 5.3 Flashprompt; selected context (shared peer answers, forwarded snippets, web sources); generation parameters; technical metadataStated policy2026-08-27

2. Platform features that send your content

These features run models you did not pick. Each sends the content it needs to the host shown; where a fallback is configured the row reads primary → fallback, and the fallback receives the same content when the primary is unavailable. Every host named here has a row in section 1.

FeatureHosts
Web search: deciding whether to searchOpenAI → Google
Web search: query optimisationOpenAI → Fireworks
Web search: source verificationGoogle → OpenAI
Round takeawayOpenAI → Fireworks
Session takeaway (when you share a session)Fireworks
Session titleOpenAI

3. Subprocessors

Service providers that process data for the functions needed to operate mumo. Each processes data only on our instructions and only for the function it provides.

ProviderFunctionReceives
VercelHosting and edge deliveryconversation content in transit; account and billing details
SupabaseDatabase and authenticationconversation content at rest; account and billing details
PostHogProduct analyticsinteraction events — all text and inputs masked
Brave SearchWeb searchsearch queries derived from your prompt

Generated from live routing configuration on 2026-08-28. Provider commitments re-checked quarterly and before any routing change.