Struct ExtractionConfig 

pub struct ExtractionConfig {

    pub provider: LlmProvider,
    pub api_key: Option<String>,
    pub api_url: String,
    pub model: String,
    pub region: Option<String>,
    pub max_extractions_per_conversation: usize,
    pub quality_threshold: f32,
    pub deduplication_threshold: f32,
    pub enable_contradiction_check: bool,
    pub enable_deduplication: bool,
    pub extraction_passes: usize,
    pub max_output_tokens: u32,
    pub effort: Option<String>,
    pub reader_model: Option<String>,
    pub request_timeout_secs: u64,
    pub connect_timeout_secs: u64,
}

Configuration for the extraction pipeline.

Fields

provider: LlmProvider

Which LLM provider to use.

api_key: Option<String>

API key for the provider (not needed for Ollama).

api_url: String

API endpoint URL (defaults based on provider).

Ignored for the Bedrock provider, whose endpoint is derived from region and model per-request.

model: String

Model name to use.

region: Option<String>

AWS region for the Bedrock provider (e.g. “us-east-1”). Ignored by all other providers. Defaults from MENTEDB_LLM_REGION/AWS_REGION.

max_extractions_per_conversation: usize

Maximum number of memories to extract from a single conversation.

quality_threshold: f32

Minimum confidence score for a memory to be accepted (0.0 to 1.0).

deduplication_threshold: f32

Embedding similarity above which a memory is considered a duplicate (0.0 to 1.0).

enable_contradiction_check: bool

Whether to check new memories against existing ones for contradictions.

enable_deduplication: bool

Whether to check new memories against existing ones for duplicates.

extraction_passes: usize

Number of extraction passes (1 = single pass, 2 = first pass + verification).

max_output_tokens: u32

Output token ceiling for each LLM call. Claude 5.5 models think before answering by default and the thinking counts against this budget, so it must leave room for both. Only tokens actually produced are billed.

effort: Option<String>

Reasoning effort for Claude models, sent as output_config.effort (“low”, “medium”, “high”). None keeps the model default. Lower effort thinks less, which is faster and cheaper.

reader_model: Option<String>

Model for judge and rerank calls (reading and comparing memories), when it should differ from model, which stays the extraction and enrichment model. None uses model for both.

request_timeout_secs: u64

Whole request timeout for one LLM call, in seconds.

connect_timeout_secs: u64

Connection timeout for one LLM call, in seconds.

Implementations

impl ExtractionConfig

pub fn with_output_env(self) -> Self

Apply MENTEDB_LLM_MAX_OUTPUT_TOKENS and MENTEDB_LLM_EFFORT from the environment, keeping the current values when they are unset or invalid.

pub fn reader_config(&self) -> Self

The config for judge and rerank calls: this one with model set to reader_model when that is given.

pub fn openai(api_key: impl Into<String>) -> Self

Create a config for OpenAI with the given API key.

pub fn anthropic(api_key: impl Into<String>) -> Self

Create a config for Anthropic with the given API key.

pub fn ollama() -> Self

Create a config for a local Ollama instance.

pub fn bedrock(region: impl Into<String>) -> Self

Create a config for AWS Bedrock (Anthropic Claude on Bedrock) in the given region. Credentials are read from the AWS environment variables at call time, not stored here. Requires the bedrock feature to actually dispatch calls.

Trait Implementations

impl Clone for ExtractionConfig

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for ExtractionConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for ExtractionConfig

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for ExtractionConfig

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for ExtractionConfig

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.