Skip to content
SilktideHelp

AI models and providers

Silktide names the AI it uses. There is no Silktide-branded abstraction over the top: you can see which company processes your data and which model answered a question, because that is what a compliance review needs and what a curious user wants.

Two things are chosen separately: providers (who may process your data) and models (which one answers).

Providers Silktide currently offers

A provider is the company that runs the model. Enabling one permits it to process your account's data as a , and nothing runs on a provider you have not enabled.

Your account administrators choose this on the AI screen: a region first (United States or European Union), then providers from the set that region offers. Each provider shows its contracting entity and a link to that company's data processing addendum.

The five providers Silktide offers today:

The choice applies to everyone working in your account, including Silktide's own staff. We cannot use a provider you have not enabled.

Models Silktide currently offers

Within the providers you allow, each offers one or more models. You only see models from providers your account has enabled.

Silktide's current recommendation (the default for new conversations unless an administrator pins something else) is Grok 4.7 from SpaceXAI. That recommendation changes as newer models prove themselves.

OpenAI

GPT-6 Sol: OpenAI's current generation, built for complex coding and multi-step agent work. This is the default OpenAI model when that provider is enabled.

GPT-6 Luna: OpenAI's most efficient GPT-6 model, for quick, focused tasks.

GPT-5.6: The previous GPT generation, kept available so you can stay on a model you have already validated.

GPT-5.5: An older GPT generation, kept available for the same reason.

Anthropic

Claude Sonnet 5: Anthropic's current Sonnet model, a balance of speed and intelligence for coding, tool use, and everyday professional work. This is the default Anthropic model when that provider is enabled.

Claude Fable 5.1: Anthropic's most capable generally available model, for the hardest, longest-running work.

Claude Fable 5: The previous Fable model, kept available so you can stay on a model you have already validated.

SpaceXAI

Grok 4.7: SpaceXAI's latest model, built for coding, long-running agent work, and knowledge work. This is Silktide's current recommendation.

Grok 4.6: The previous Grok generation, kept available so you can stay on a model you have already validated.

Grok 4.5: An older Grok generation, kept available for the same reason.

Google

Gemini 3.8 Flash: Google's newest Flash model, for fast answers and agent work. This is the default Google model when that provider is enabled.

Gemini 3.6 Flash: An earlier Flash model, kept available so you can stay on a model you have already validated.

Mistral

Mistral Medium 3.5: Mistral's current flagship for agent work. This is the default Mistral model when that provider is enabled.

Mistral Large 3: An open-weight workhorse kept available alongside Medium.

How a model is chosen

Most people never need to change this: an administrator sets a default model for the account, and conversations start there.

Administrators can either pin a specific model, or leave the account on Silktide's recommendation. On the recommendation, your account moves to newer models as they prove themselves. Pin a specific model when you need to be able to state exactly which one is in use.

If you want to change the model for a single conversation, open Model in the Agent's message box. That row is only there when your account has more than one model to choose between.

Backup model

If the default model is slow to produce a reply, a conversation can fail over to a backup model partway through, so you are not left waiting on a provider having a bad moment. The backup is set on the AI screen, under Default model: leave it on Silktide's recommendation, pin a specific model, or turn it off with No backup. Only models from providers you have enabled are offered, in the same AI region as the rest of your account. A reply that used the backup is marked Backup in the conversation, naming the model that was too slow, so it is always clear which model actually answered.

Thinking effort

Separately from which model runs, you choose how hard it thinks. Lower effort answers quickly and costs less; higher effort reasons for longer and is better on genuinely difficult problems. Set it with the slider in the Agent's message box, per conversation.

The available levels come from the provider, not from Silktide, so the names differ between models. OpenAI models offer None, Low, Medium, High, and Extra high. SpaceXAI, Anthropic, and Google models offer Low, Medium, and High. Mistral Medium offers None and High. Silktide shows whatever the model you are using actually offers rather than inventing its own scale, which means a level always means what the provider says it means.

Some parts of Silktide set the effort themselves. An AI report that has to write something original asks for the highest effort available, because the result is worth the wait.

Why models change

Model quality, speed, price, and availability all move quickly, and we expect to add and retire models several times a year. Naming providers and models honestly is compatible with that: the provider you consented to is a stable commitment, while which of that provider's models we recommend is not.

Retiring a model never changes which providers may process your data. That only changes when an administrator changes it.

What is sent to a provider

Only the context needed for the requested work. For Agent conversations, that can include your prompt, relevant conversation history, permitted account or website context, information retrieved for your request, and any attachments you chose to provide.

A conversation's title is generated using the same model and provider as the conversation itself, so naming a thread cannot send its contents anywhere you have not agreed to.

A few features use a model Silktide chooses rather than the providers you enabled, because they do not send your content: spell check verification and report illustrations. Both are listed on the AI screen.

See Data, privacy, and retention for the distinction between data Silktide holds and data processed by a provider.

Last updated

Was this page helpful?

AI models and providers | Silktide Help