Convert causal relations from LLM format to ExtractedFact format. Adjusts target_fact_index from content-relative to global indices.

Convert causal relations from LLM format to ExtractedFact format. Adjusts target_fact_index from content-relative to global indices.

Convert RetrievalResult seeds to SeedNode format.

Create a new entity or get existing one if it already exists. Uses INSERT ... ON CONFLICT to handle race conditions where tw

Validate vector extension: 'pgvector', 'vchord', or 'pgvectorscale'. Args: conn: SQLAlchemy connection object vector_extensi

Extract a header value from ASGI scope.

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (e.g., 'tenant_x.' or '' for public).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (e.g., 'tenant_x.' or '' for public).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (e.g., 'tenant_x.' or '' for public).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (required for multi-tenant support).

Get schema prefix for table names (e.g., 'tenant_x.' or '' for public).

Get schema prefix for table names (required for multi-tenant support).

Get the target schema name (tenant schema or 'public').

Get the target schema name (tenant schema or 'public').

Deterministic, schema-safe HNSW index name for a (bank, fact_type) pair. Uses the first 16 hex chars of internal_id (8 bytes of entropy) — unique

Infer a temporal date from fact text when LLM didn't provide occurred_start. This is a fallback for when the LLM fails to extract temporal i

Infer a temporal date from fact text when LLM didn't provide occurred_start. This is a fallback for when the LLM fails to extract temporal i

For label groups with tag=True, add extracted key:value label entities to each fact's tags list. Modifies facts in place. This lets enti

For label groups with tag=True, add extracted key:value label entities to each fact's tags list. Modifies facts in place. This lets enti

Interpolate between two RGB colors.

Check if the tool name represents the 'done' tool.

Load and cache the AI chat history index.json.

Return the maximum non-None datetime from an iterable.

Return the maximum non-None datetime from an iterable.

Migrate legacy free_values/multi_value fields to the new type field.

Migrate the embedding column of a single table to the required dimension. - If dimensions match: no action needed - If dimensions differ

Return the minimum non-None datetime from an iterable.

Return the minimum non-None datetime from an iterable.

Parse http_config column value (JSONB returned as text or dict) into a model.

Parse a comma-separated string into a non-empty list of stripped tokens.

Process a batch of memories in a single LLM call. Steps: 1. Parallel recalls — one per fact (read-only; safe to parallelise) 2. Unio

Process a batch of memories in a single LLM call. Steps: 1. Parallel recalls — one per fact (read-only; safe to parallelise) 2. Unio

Shared scoring + upsert logic used by both lookup strategies.

Normalize parser value to a non-empty list of names.

Internal method for batch processing without chunking logic. Assumes contents are already appropriately sized (< 50k chars).

Convert a database row to an observation dict.

Convert LLMToolCall to OpenAI message format.

Truncate text to at most max_tokens using the shared tiktoken encoder.

Persist last HTTP attempt info into async_operations.result_metadata.

Raise HTTP 400 if any parser name is not in the allowlist.

Add loaded edges to the cache (all edge types at once). Args: edges_by_type: Dict mapping edge_type -> from_node_id -> l

No-op.

Authenticate the action context and return tenant context. Args: context: The action context containing API key and othe

Build a content dict for retain operations. Args: content: The memory content context: Category for the memory timestamp:

Build a dynamic Pydantic model for structured label extraction. Each LabelGroup becomes a typed field based on its type: - type="text"

Split text into chunks, preserving conversation structure when possible. For JSON conversation arrays (user/assistant turns), splits at turn

Split text into chunks, preserving conversation structure when possible. For JSON conversation arrays (user/assistant turns), splits at turn

Create and set the global metrics collector. Should be called after initialize_metrics().

Delete a single memory unit and all its associated links. Due to CASCADE DELETE constraints, this will automatically delete:

Ensure the embedding column dimension matches the model's dimension for all tables. Checks and adjusts memory_units.embedding and mental_mod

Ensure the text search columns and indexes match the configured extension. This function checks the current search_vector column type and in

Ensure the vector indexes match the configured vector extension. This function checks the current vector index type in the database and

Extract semantic facts from conversational or narrative text using LLM. For large texts (>3000 chars), automatically chunks at sentence boun

Queue webhook deliveries within an existing database connection/transaction. Identical to fire_event() but uses the provided connect

Flush accumulated entity stats and co-occurrence counts for the current task. Must be called AFTER the retain transaction commits.

Format facts as JSON for LLM prompt.

Create ProcessedFact from ExtractedFact. Args: extracted_fact: Source ExtractedFact embedding: Generated emb

Get set of config fields that this tenant/bank is allowed to modify. This method controls which configurable fields can be modified

Return only the config fields that were explicitly set. reflect_mission takes precedence over deprecated mission/background aliases.

Get and process the database URL from config or environment. Returns the URL with the correct driver (psycopg2) for migrations.

Get or create the default graph retriever based on config.

Get the list of recorded mock calls.

Get the Python logging level from the configured log level string.

Convert a token count to a bucket label for use as a dimension. This allows analyzing token usage patterns without high-cardinality issues.

Find a visit by node ID.

Perform a health check by querying the database. Returns: dict with status and optional error message Note: