Chat and queries

Choosing a model

The selectable models, what each is for, and what switching one does to your thread.

ChatPlotDB is not built on a single model. Several are selectable, each suited to a different job, and the routing between providers is handled for you.

The lineup

ModelVendorUse it for
Claude Sonnet 5AnthropicThe default: schema reasoning and SQL at Sonnet cost
Claude Opus 4.8AnthropicThe deepest reasoning; also the Config Studio default
Claude Sonnet 4.6AnthropicThe previous default, same tier: a second opinion
Claude Haiku 4.5AnthropicFastest and cheapest; the guest default
GPT-5.5OpenAITeams standardised on OpenAI, or cross-vendor comparison
GLM 5.2Z.aiOpen weights, roughly half the input cost of a Sonnet turn
DeepSeek V4 ProDeepSeekStrong reasoning at a low rate for high-volume analysis

Guests get a reduced set: Haiku 4.5 and both Sonnets. Requesting anything outside that set as a guest is silently downgraded rather than refused.

Switching model starts a new thread

A conversation is pinned server-side to the model that answered it. Picking a different model opens a fresh thread; the previous one keeps its place in your history.

Your connected database and its config carry straight over: those belong to the connection, not the conversation.

Querying the list

The selectable set is a server fact, gated by whether you are signed in and which provider keys the deployment has:

GET /api/models

It returns the chat and config model sets, each with its default and a provider id per model.

A missing key degrades, it does not fail

If a deployment has no key for the model you picked, the turn quietly falls back to the default instead of erroring.

On this page