Interface: IModelRequest

Defined in: internal/llm-client/types.ts:91

One request to a model. Built by ContextAssembler.

Properties

cacheTools?

optional cacheTools?: boolean

Defined in: internal/llm-client/types.ts:113

Puts the provider's cache marker on the last tool definition too (Anthropic).

A read point for the tool definitions alone, for calls whose system blocks differ but whose tools do not.

Set by the ContextAssembler only when the prompt cache may be shared across tenants: tool definitions are the same for every tenant, so a tools-only entry is always shared.

Default

false

---

maxOutputTokens

maxOutputTokens: number

Defined in: internal/llm-client/types.ts:119

The output cap of the call: the profile's, set by the ContextAssembler; the output of the usage check.

---

messages

messages: IMessage[]

Defined in: internal/llm-client/types.ts:115

The transcript the model continues: the history, the user turn, the tool turns of a run.

---

modelId

modelId: string

Defined in: internal/llm-client/types.ts:93

The registry id of the model to call.

---

outputSchema?

optional outputSchema?: ZodType<unknown, unknown, $ZodTypeInternals<unknown, unknown>>

Defined in: internal/llm-client/types.ts:117

Zod schema for structured output; omitted = free text.

---

system

system: ISystemBlock[]

Defined in: internal/llm-client/types.ts:102

System prompt as blocks, in order (rendered CHAT_MAIN: the stable block, then the per-call block).

Anthropic gets one system block per entry, marked where cache is set; the others get the texts joined (joinSystem), OpenAI with a cache key derived from the cache block.

Example

[{text: "You are the assistant…", cache: true}, {text: "Today is Thursday, 2026-10-29…"}]

---

temperature?

optional temperature?: number

Defined in: internal/llm-client/types.ts:124

Sampling temperature, in the provider's range (OpenAI 0–2, Anthropic 0–1); omit for the provider default. Only on a model whose caps.temperature is true: the client refuses it otherwise (400, D82).