CacheConfig
Defined in: src/models/model.ts:72
Configuration for prompt caching.
Providers consume only the fields they support.
Properties
Section titled “Properties”strategy?
Section titled “strategy?”optional strategy?: "auto" | "anthropic";Defined in: src/models/model.ts:81
Whether to skip caching for models that do not support it.
- “auto”: cache only when the model is known to support it
- “anthropic”: cache without that check, for model identifiers it cannot inspect (an application inference profile, for example)
Default Value
Section titled “Default Value”'auto'optional ttl?: CacheTTL;Defined in: src/models/model.ts:93
TTL for every cache point, overridden by a per-section TTL. Provider default when omitted.
Bedrock requires checkpoint TTLs to be non-increasing across toolConfig, system and messages, and
rejects a longer TTL that follows a shorter one. This TTL therefore also fills in for a cache point
placed by hand in the system prompt that carries none of its own, so one value keeps every
checkpoint in step. A TTL written on such a point is left as written, and a toolsTTL that differs
from this one leaves the point at the provider default rather than landing a longer TTL behind a
shorter checkpoint - either way, two TTLs in tension are yours to reconcile.
toolsTTL?
Section titled “toolsTTL?”optional toolsTTL?: boolean | CacheTTL;Defined in: src/models/model.ts:100
Cache the tool definitions. A TTL sets this section’s duration; false disables it.
Default Value
Section titled “Default Value”truesystemPromptTTL?
Section titled “systemPromptTTL?”optional systemPromptTTL?: boolean | CacheTTL;Defined in: src/models/model.ts:109
Cache the system prompt, auto-injecting a cache point at its end so repeated calls with the same
static system prefix hit the cache. A TTL sets this section’s duration; true (the default) reads
the value from ttl; false disables systemPrompt cache injection.
Default Value
Section titled “Default Value”truemessagesTTL?
Section titled “messagesTTL?”optional messagesTTL?: boolean | CacheTTL;Defined in: src/models/model.ts:117
Cache the conversation prefix, on the last user message. A TTL sets this section’s duration;
false disables it.
Default Value
Section titled “Default Value”truecacheKey?
Section titled “cacheKey?”optional cacheKey?: string;Defined in: src/models/model.ts:120
Stable identity a provider can use to route its cache.