API Reference¶

Auto-generated from source code docstrings. For conceptual guides on how these components work together, see the Concepts section.

Client¶

MemoryClient¶

MemoryClient ¶

Main entry point for the arandu SDK.

Usage::

from arandu import MemoryClient, MemoryConfig
from arandu.providers.openai import OpenAIProvider

provider = OpenAIProvider(api_key="sk-...")
memory = MemoryClient(
    database_url="postgresql+psycopg://user:pass@localhost/mydb",
    llm=provider,
    embeddings=provider,
)
await memory.initialize()

result = await memory.write(agent_id="my_agent", message="I live in São Paulo", speaker_name="Pedro")
context = await memory.retrieve(agent_id="my_agent", query="where does Pedro live?")

initialize `async` ¶

initialize() -> None

Create memory tables in the database (idempotent).

write `async` ¶

write(agent_id: str, message: str, speaker_name: str, source: str = 'api', *, session_id: str = 'default', config_overrides: dict | None = None, dry_run: bool = False, recent_messages: list[str] | None = None, verbose: bool = False, occurred_at: datetime | None = None) -> WriteResult

Extract facts from a message and store them in memory.

Parameters:

Name	Type	Description	Default
`agent_id`	`str`	Identifier for the agent whose memory this is. All facts, entities, and relationships are scoped to this agent. Use the same agent_id to share memory across sessions.	required
`message`	`str`	The message text to extract knowledge from.	required
`speaker_name`	`str`	Name of the speaker (the person talking). Required. This identifies WHO is speaking — the person whose words are being memorized. First-person pronouns ("I", "me", "my") resolve to this name. Example: `speaker_name="Pedro Menezes"`.	required
`source`	`str`	Source channel identifier (default: "api").	`'api'`
`session_id`	`str`	Identifier for the conversation session (default: "default"). Use different session_ids to separate conversation contexts. The agent's memory persists across sessions, but session_id helps track which conversation each event came from.	`'default'`
`config_overrides`	`dict \| None`	Optional dict of MemoryConfig field overrides for this request. Only the provided keys are changed; all others inherit from the client config. Example: `{"extraction_timeout_sec": 60.0}`	`None`
`dry_run`	`bool`	If True, run extraction + resolution + reconciliation but skip the upsert stage. No data is persisted. Useful for benchmarking.	`False`
`recent_messages`	`list[str] \| None`	Optional conversation context (last N messages) for resolving pronouns and anaphora references in extraction.	`None`
`verbose`	`bool`	If True, include pipeline trace with intermediate data.	`False`
`occurred_at`	`datetime \| None`	Optional timestamp for when this message was sent. Defaults to now. Use for importing historical conversations so the extraction can resolve relative dates correctly.	`None`

Returns:

Type	Description
`WriteResult`	WriteResult with details of extracted and stored facts.

Raises:

Type	Description
`ValueError`	If speaker_name is empty or whitespace.

retrieve `async` ¶

retrieve(agent_id: str, query: str, *, session_id: str | None = None, config_overrides: dict | None = None, verbose: bool = False, entity_keys: list[str] | None = None) -> RetrieveResult

Retrieve relevant memory context for a query.

Parameters:

Name	Type	Description	Default
`agent_id`	`str`	Identifier for the agent whose memory to search.	required
`query`	`str`	The query to search memory for.	required
`session_id`	`str \| None`	Optional session filter. If None (default), searches across all sessions. If provided, may be used for context prioritization in future versions.	`None`
`config_overrides`	`dict \| None`	Optional dict of MemoryConfig field overrides for this request. Only the provided keys are changed; all others inherit from the client config. Example: `{"enable_reranker": False, "topk_facts": 10}`	`None`
`verbose`	`bool`	If True, include pipeline trace with intermediate data.	`False`
`entity_keys`	`list[str] \| None`	Optional list of entity keys to filter by (OR logic). When provided, only facts linked to at least one of the specified entities are returned. Example: `["person:pedro", "person:ana"]`	`None`

Returns:

Type	Description
`RetrieveResult`	RetrieveResult with ranked facts and formatted context string.

get `async` ¶

get(agent_id: str, fact_id: str) -> FactDetail | None

Return a single fact by ID, or None if not found.

Fetches any fact belonging to agent_id regardless of its valid_to status — even facts that have been soft-deleted by the reconciliation pipeline are returned when looked up by ID.

Parameters:

Name	Type	Description	Default
`agent_id`	`str`	Agent that owns the fact.	required
`fact_id`	`str`	UUID of the fact to retrieve.	required

Returns:

Type	Description
`FactDetail \| None`	A `FactDetail` with the fact's metadata, or `None`
`FactDetail \| None`	if no matching fact exists.

get_all `async` ¶

get_all(agent_id: str, *, limit: int = 100, offset: int = 0, entity_keys: list[str] | None = None) -> list[FactDetail]

List active facts for an agent with pagination.

Only facts with valid_to IS NULL (i.e. not soft-deleted by the reconciliation pipeline) are returned. Results are ordered by created_at descending (newest first).

Parameters:

Name	Type	Description	Default
`agent_id`	`str`	Agent whose facts to list.	required
`limit`	`int`	Maximum number of facts to return (default 100).	`100`
`offset`	`int`	Number of facts to skip (default 0).	`0`
`entity_keys`	`list[str] \| None`	Optional list of entity keys to filter by (OR logic). When provided, only facts linked to at least one of the specified entities are returned. Aliases registered in `memory_entity_aliases` are resolved automatically to their canonical key before filtering. Keys that resolve to nothing are silently dropped from the filter (the CRUD path does not return warnings — use `retrieve()` for that).	`None`

Returns:

Type	Description
`list[FactDetail]`	A list of `FactDetail` objects.

delete `async` ¶

delete(agent_id: str, fact_id: str) -> bool

Hard-delete a single fact by ID.

The row is physically removed from the database. Associated MemoryFactEntityLink rows are removed automatically via the CASCADE foreign-key constraint.

Parameters:

Name	Type	Description	Default
`agent_id`	`str`	Agent that owns the fact.	required
`fact_id`	`str`	UUID of the fact to delete.	required

Returns:

Type	Description
`bool`	`True` if a fact was deleted, `False` if no matching
`bool`	fact was found.

delete_all `async` ¶

delete_all(agent_id: str) -> int

Hard-delete all memory for an agent.

Removes all facts, entities, aliases, and relationships belonging to agent_id. MemoryFactEntityLink rows are removed automatically via the CASCADE foreign-key constraint on MemoryFact. MemoryEvent rows are preserved (immutable audit log).

Parameters:

Name	Type	Description	Default
`agent_id`	`str`	Agent whose memory to delete.	required

Returns:

Type	Description
`int`	The number of facts deleted.

entities `async` ¶

entities(agent_id: str, *, limit: int = 100, offset: int = 0) -> list[EntityDetail]

List known entities for an agent.

Returns active entities (is_active = True) ordered by last_seen_at descending (most recently seen first), with id as tiebreaker for stable pagination.

Parameters:

Name	Type	Description	Default
`agent_id`	`str`	Agent whose entities to list.	required
`limit`	`int`	Maximum number of entities to return (default 100).	`100`
`offset`	`int`	Number of entities to skip (default 0).	`0`

Returns:

Type	Description
`list[EntityDetail]`	A list of `EntityDetail` objects.

close `async` ¶

close() -> None

Dispose of the database engine and release connections.

WriteResult¶

WriteResult `dataclass` ¶

Result of a memory write operation.

RetrieveResult¶

RetrieveResult `dataclass` ¶

Result of a memory retrieve operation.

The warnings field contains user-facing strings for any entity_keys the caller passed that did not resolve to a canonical entity or alias. When the filter matches fully, warnings is empty.

FactDetail¶

FactDetail `dataclass` ¶

Detailed view of a stored fact — used by CRUD operations (get, get_all).

Unlike ScoredFact (which carries retrieval scores), FactDetail exposes the full metadata of a fact as stored in the database.

EntityDetail¶

EntityDetail `dataclass` ¶

Detailed view of a known entity — used by the entities() method.

ScoredFact¶

ScoredFact `dataclass` ¶

A fact with retrieval scores.

The scores dict contains the breakdown of individual scoring signals. Possible keys (not all are always present):

semantic — cosine similarity from pgvector (0.0–1.0)
keyword — fraction of query words found in fact text (0.0–1.0)
recency — exponential decay based on fact age (configurable half-life)
importance — base importance value of the fact (typically 0.5)
confidence — effective confidence with temporal decay
pattern — additive boost for reinforced facts (up to 0.1)
graph — score from BFS 2-hop knowledge graph traversal
reranker — LLM-based relevance score (0.0–1.0, only when reranker is enabled)

PipelineTrace¶

PipelineTrace `dataclass` ¶

Collects intermediate data from pipeline stages.

Usage in pipeline functions::

trace = PipelineTrace()
t0 = trace.start()
# ... run extraction ...
trace.add_step("extraction", {"entities": [...]}, trace.elapsed(t0))

start ¶

start() -> float

Return a perf_counter timestamp for measuring step duration.

elapsed ¶

elapsed(t0: float) -> float

Return elapsed milliseconds since t0.

add_step ¶

add_step(name: str, data: dict, duration_ms: float) -> None

Record a pipeline step with its data and timing.

to_dict ¶

to_dict() -> dict

Serialize to a JSON-compatible dict.

Configuration¶

MemoryConfig¶

MemoryConfig `dataclass` ¶

Configuration for the memory system.

All fields have sensible defaults. Override only what you need.

Protocols¶

LLMProvider¶

LLMProvider ¶

Bases: Protocol

Protocol for LLM completion calls.

complete `async` ¶

complete(messages: list[dict], temperature: float = 0, response_format: dict | None = None, max_tokens: int | None = None) -> LLMResult

Send a chat completion request and return the result.

Parameters:

Name	Type	Description	Default
`messages`	`list[dict]`	List of message dicts with 'role' and 'content' keys.	required
`temperature`	`float`	Sampling temperature (0 = deterministic).	`0`
`response_format`	`dict \| None`	Optional format spec (e.g. {"type": "json_object"}).	`None`
`max_tokens`	`int \| None`	Optional max tokens for the response.	`None`

Returns:

Type	Description
`LLMResult`	LLMResult with the response text and optional token usage.

EmbeddingProvider¶

EmbeddingProvider ¶

Bases: Protocol

Protocol for text embedding generation.

embed `async` ¶

embed(texts: list[str]) -> list[list[float]]

Generate embeddings for a batch of texts.

Parameters:

Name	Type	Description	Default
`texts`	`list[str]`	List of text strings to embed.	required

Returns:

Type	Description
`list[list[float]]`	List of embedding vectors (one per input text).

embed_one `async` ¶

embed_one(text: str) -> list[float] | None

Generate an embedding for a single text.

Parameters:

Name	Type	Description	Default
`text`	`str`	Text string to embed.	required

Returns:

Type	Description
`list[float] \| None`	Embedding vector, or None if the text is empty/invalid.

Providers¶

OpenAIProvider¶

OpenAIProvider ¶

OpenAI implementation of LLMProvider + EmbeddingProvider.

Implements both protocols with a single class via composition (not inheritance). The openai SDK is imported lazily to keep it an optional dependency.

Usage::

from arandu.providers.openai import OpenAIProvider

provider = OpenAIProvider(api_key="sk-...")
memory = MemoryClient(
    database_url="postgresql+psycopg://...",
    llm=provider,
    embeddings=provider,
)

complete `async` ¶

complete(messages: list[dict], temperature: float = 0, response_format: dict | None = None, max_tokens: int | None = None) -> LLMResult

Send a chat completion request and return the result with usage.

embed `async` ¶

embed(texts: list[str]) -> list[list[float]]

Generate embeddings for a batch of texts.

embed_one `async` ¶

embed_one(text: str) -> list[float] | None

Generate an embedding for a single text.

AnthropicProvider¶

Built-in LLM provider for Anthropic Claude models. Implements LLMProvider only - does not provide embeddings (use OpenAIProvider or another embedding provider alongside).

from arandu.providers.anthropic import AnthropicProvider

llm = AnthropicProvider(
    api_key="sk-ant-...",       # Anthropic API key
    model="claude-sonnet-4-20250514",  # Default model
    timeout=30.0,               # Request timeout in seconds
    max_tokens=4096,            # Default max tokens per response
)

Install: pip install arandu[anthropic]

Key behaviors:

System messages are extracted from the messages list and passed via Anthropic's system parameter
response_format={"type": "json_object"} appends a JSON instruction to the system prompt (Anthropic doesn't support this natively)
Token usage is tracked via LLMResult.usage
Markdown code fences are stripped automatically

Exceptions¶

All SDK exceptions inherit from MemoryError. The SDK is fail-safe by default - most errors are caught internally and result in graceful degradation (empty results, logged warnings). These exceptions are raised only when explicitly documented.

MemoryError (base)
├── ExtractionError    — LLM extraction failed
├── ResolutionError    — entity resolution failed
├── ReconciliationError — fact reconciliation failed
├── RetrievalError     — memory retrieval failed
└── UpsertError        — database persistence failed

MemoryError¶

Base exception for all arandu SDK errors. Catch this for catch-all handling.

from arandu import MemoryError

try:
    result = await memory.write(agent_id, message, speaker_name="Rafael")
except MemoryError as e:
    print(f"SDK error: {e}")
    print(f"Caused by: {e.__cause__}")  # original exception, if any

ExtractionError¶

Raised when fact/entity extraction from a message fails (LLM timeout, invalid JSON response, rate limit). In practice, the write pipeline catches this internally and returns an empty WriteResult with success=True - the event is still logged.

ResolutionError¶

Raised when entity resolution fails (embedding provider down, LLM disambiguation timeout). The pipeline catches this internally - unresolved entities are created as new entities.

ReconciliationError¶

Raised when fact reconciliation against existing knowledge fails (LLM call fails for similarity evaluation). The pipeline catches this internally - defaults to ADD (better to have a near-duplicate than lose information).

RetrievalError¶

Raised when memory retrieval fails (semantic search error, reranker timeout). The read pipeline catches this internally - returns empty results rather than crashing.

UpsertError¶

Raised when upserting facts or entities into the database fails (constraint violation, connection error). Individual fact upserts use savepoints - if one fact fails, the others proceed normally.

Error handling example¶

from arandu import MemoryError, ExtractionError, RetrievalError

# Write — fail-safe by default
result = await memory.write(agent_id, message, speaker_name="Rafael")
if not result.success:
    print(f"Write failed: {result.error}")
# No try/except needed — the pipeline handles errors internally

# Retrieve — also fail-safe
result = await memory.retrieve(agent_id, query)
# Empty facts = nothing found OR error occurred
# Check result.duration_ms to detect timeouts

Background Functions¶

Clustering¶

cluster_user_facts¶

cluster_user_facts `async` ¶

cluster_user_facts(session: AsyncSession, agent_id: str, embedding_provider: EmbeddingProvider, llm_provider: LLMProvider, config: MemoryConfig) -> ClusteringResult

Build or update clusters for a user's facts. Idempotent.

Groups active facts by (entity_type, entity_key), creates or updates clusters with LLM-generated summaries and embeddings.

Parameters:

Name	Type	Description	Default
`session`	`AsyncSession`	Database session.	required
`agent_id`	`str`	User identifier.	required
`embedding_provider`	`EmbeddingProvider`	Injected embedding provider.	required
`llm_provider`	`LLMProvider`	Injected LLM provider.	required
`config`	`MemoryConfig`	Memory configuration.	required

Returns:

Type	Description
`ClusteringResult`	ClusteringResult with stats.

detect_communities¶

detect_communities `async` ¶

detect_communities(session: AsyncSession, agent_id: str, embedding_provider: EmbeddingProvider, llm_provider: LLMProvider, config: MemoryConfig) -> CommunityDetectionResult

Find cross-entity themes by comparing cluster embeddings.

Uses Union-Find to group clusters from different entity_keys with similar embeddings into communities, materialized as MemoryMetaObservation.

Parameters:

Name	Type	Description	Default
`session`	`AsyncSession`	Database session.	required
`agent_id`	`str`	User identifier.	required
`embedding_provider`	`EmbeddingProvider`	Injected embedding provider.	required
`llm_provider`	`LLMProvider`	Injected LLM provider.	required
`config`	`MemoryConfig`	Memory configuration.	required

Returns:

Type	Description
`CommunityDetectionResult`	CommunityDetectionResult with stats.

Consolidation¶

run_consolidation¶

run_consolidation `async` ¶

run_consolidation(session: AsyncSession, agent_id: str, llm_provider: LLMProvider, config: MemoryConfig, *, embedding_provider: EmbeddingProvider | None = None) -> ConsolidationResult

L2 Periodic consolidation: detect patterns from events and facts.

Loads events within the configured lookback window, detects patterns via LLM, and creates/reinforces meta-observations with dedup.

Parameters:

Name	Type	Description	Default
`session`	`AsyncSession`	Database session.	required
`agent_id`	`str`	User identifier.	required
`llm_provider`	`LLMProvider`	Injected LLM provider.	required
`config`	`MemoryConfig`	Memory configuration.	required
`embedding_provider`	`EmbeddingProvider \| None`	Optional embedding provider for dedup.	`None`

Returns:

Type	Description
`ConsolidationResult`	ConsolidationResult with stats.

run_profile_consolidation¶

run_profile_consolidation `async` ¶

run_profile_consolidation(session: AsyncSession, agent_id: str, llm_provider: LLMProvider) -> dict[str, Any]

L3 Profile consolidation: refresh entity summaries via LLM.

For each active MemoryEntity, fetches active facts and generates a summary. Caps at 20 entities per run.

Parameters:

Name	Type	Description	Default
`session`	`AsyncSession`	Database session.	required
`agent_id`	`str`	User identifier.	required
`llm_provider`	`LLMProvider`	Injected LLM provider.	required

Returns:

Type	Description
`dict[str, Any]`	Dict with `entities_processed`, `summaries_generated`, `summaries_failed`.

Memify¶

run_memify¶

run_memify `async` ¶

run_memify(session: AsyncSession, agent_id: str, llm_provider: LLMProvider, embedding_provider: EmbeddingProvider, config: MemoryConfig) -> MemifyResult

Run the memify pipeline: vitality scoring → staleness → edge management.

Step 1: Score all active facts for vitality; mark stale if below threshold. Step 2: Reinforce/decay KG edge strengths based on recency of use. Step 3: (Optional) Detect and execute entity merges.

Parameters:

Name	Type	Description	Default
`session`	`AsyncSession`	Database session.	required
`agent_id`	`str`	User identifier.	required
`llm_provider`	`LLMProvider`	Injected LLM provider.	required
`embedding_provider`	`EmbeddingProvider`	Injected embedding provider.	required
`config`	`MemoryConfig`	Memory configuration.	required

Returns:

Type	Description
`MemifyResult`	MemifyResult with pipeline stats.

compute_vitality¶

compute_vitality ¶

compute_vitality(fact: object, now: datetime | None = None, *, weights: dict[str, float] | None = None) -> float

Compute vitality score for a single fact (0.0-1.0).

Components: - retrieval (0.30): log-scale of times_retrieved - recency (0.25): exponential decay from last_retrieved_at - confidence (0.20): raw confidence value - reinforcement (0.15): bounded reinforcement count

Correction penalty: 0.8 ^ user_correction_count. Superseded facts (valid_to is not None) return 0.0.

Parameters:

Name	Type	Description	Default
`fact`	`object`	A MemoryFact-like object.	required
`now`	`datetime \| None`	Current timestamp (defaults to UTC now).	`None`
`weights`	`dict[str, float] \| None`	Optional custom weight dict.	`None`

Returns:

Type	Description
`float`	Vitality score between 0.0 and 1.0.

Sleep-Time Compute¶

compute_entity_importance¶

compute_entity_importance `async` ¶

compute_entity_importance(session: AsyncSession, agent_id: str, config: MemoryConfig) -> EntityImportanceResult

Compute importance scores for all active entities.

Formula: weighted sum of fact_density, recency, retrieval_frequency, relationship_degree. Updates MemoryEntity.importance_score in-place.

Parameters:

Name	Type	Description	Default
`session`	`AsyncSession`	Database session.	required
`agent_id`	`str`	User identifier.	required
`config`	`MemoryConfig`	Memory configuration.	required

Returns:

Type	Description
`EntityImportanceResult`	EntityImportanceResult with stats and top entities.

refresh_entity_summaries¶

refresh_entity_summaries `async` ¶

refresh_entity_summaries(session: AsyncSession, agent_id: str, llm_provider: LLMProvider, config: MemoryConfig) -> SummaryRefreshResult

Refresh entity summaries via LLM for stale entities.

An entity is stale if its summary_text is NULL or summary_refreshed_at is older than summary_refresh_interval_days.

Parameters:

Name	Type	Description	Default
`session`	`AsyncSession`	Database session.	required
`agent_id`	`str`	User identifier.	required
`llm_provider`	`LLMProvider`	Injected LLM provider.	required
`config`	`MemoryConfig`	Memory configuration.	required

Returns:

Type	Description
`SummaryRefreshResult`	SummaryRefreshResult with stats.

detect_entity_communities¶

detect_entity_communities `async` ¶

detect_entity_communities(session: AsyncSession, agent_id: str, config: MemoryConfig) -> dict[str, Any]

Connected-component clustering on entity relationship graph.

Groups entities connected by relationships with strength >= 0.3 into communities.

Parameters:

Name	Type	Description	Default
`session`	`AsyncSession`	Database session.	required
`agent_id`	`str`	User identifier.	required
`config`	`MemoryConfig`	Memory configuration.	required

Returns:

Type	Description
`dict[str, Any]`	Dict with `communities_found` and `total_entities_assigned`.

Result Dataclasses¶

ClusteringResult¶

ClusteringResult `dataclass` ¶

Result of fact clustering.

Attributes:

Name	Type	Description
`clusters_created`	`int`	Number of new clusters created.
`clusters_reinforced`	`int`	Number of existing clusters updated.
`summaries_generated`	`int`	Number of cluster summaries generated via LLM.
`facts_assigned`	`int`	Number of facts assigned to clusters.

CommunityDetectionResult¶

CommunityDetectionResult `dataclass` ¶

Result of cross-entity community detection.

Attributes:

Name	Type	Description
`communities_created`	`int`	New community observations created.
`communities_reinforced`	`int`	Existing community observations reinforced.
`clusters_in_communities`	`int`	Total clusters assigned to communities.
`skipped`	`bool`	Whether detection was skipped.
`skip_reason`	`str \| None`	Reason for skipping, if any.

ConsolidationResult¶

ConsolidationResult `dataclass` ¶

Result of L2/L3 consolidation.

Attributes:

Name	Type	Description
`events_processed`	`int`	Number of events analyzed.
`observations_created`	`int`	New meta-observations created.
`observations_reinforced`	`int`	Existing observations reinforced.
`skipped`	`bool`	Whether consolidation was skipped.
`skip_reason`	`str \| None`	Reason for skipping, if any.

MemifyResult¶

MemifyResult `dataclass` ¶

Result of the memify pipeline.

Attributes:

Name	Type	Description
`facts_scored`	`int`	Number of facts scored for vitality.
`facts_marked_stale`	`int`	Number of facts marked as stale.
`edges_reinforced`	`int`	Number of KG edges reinforced.
`merges_executed`	`int`	Number of entity merges executed.

EntityImportanceResult¶

EntityImportanceResult `dataclass` ¶

Result of entity importance scoring.

Attributes:

Name	Type	Description
`entities_scored`	`int`	Number of entities scored.
`top_entities`	`list[tuple[str, float]]`	Top entities by score (key, score) pairs.

SummaryRefreshResult¶

SummaryRefreshResult `dataclass` ¶

Result of entity summary refresh.

Attributes:

Name	Type	Description
`summaries_refreshed`	`int`	Number of summaries generated.
`summaries_skipped`	`int`	Number of entities skipped (no facts or LLM failure).

API Reference¶

Client¶

MemoryClient¶

MemoryClient ¶

initialize async ¶

write async ¶

retrieve async ¶

get async ¶

get_all async ¶

delete async ¶

delete_all async ¶

entities async ¶

close async ¶

WriteResult¶

WriteResult dataclass ¶

RetrieveResult¶

RetrieveResult dataclass ¶

FactDetail¶

FactDetail dataclass ¶

EntityDetail¶

EntityDetail dataclass ¶

ScoredFact¶

ScoredFact dataclass ¶

PipelineTrace¶

PipelineTrace dataclass ¶

start ¶

elapsed ¶

add_step ¶

to_dict ¶

Configuration¶

MemoryConfig¶

MemoryConfig dataclass ¶

Protocols¶

LLMProvider¶

LLMProvider ¶

complete async ¶

EmbeddingProvider¶

EmbeddingProvider ¶

embed async ¶

embed_one async ¶

Providers¶

OpenAIProvider¶

OpenAIProvider ¶

complete async ¶

embed async ¶

embed_one async ¶

AnthropicProvider¶

Exceptions¶

MemoryError¶

ExtractionError¶

ResolutionError¶

ReconciliationError¶

RetrievalError¶

UpsertError¶

Error handling example¶

Background Functions¶

Clustering¶

cluster_user_facts¶

cluster_user_facts async ¶

detect_communities¶

detect_communities async ¶

Consolidation¶

run_consolidation¶

run_consolidation async ¶

run_profile_consolidation¶

run_profile_consolidation async ¶

Memify¶

run_memify¶

run_memify async ¶

compute_vitality¶

compute_vitality ¶

Sleep-Time Compute¶

compute_entity_importance¶

compute_entity_importance async ¶

refresh_entity_summaries¶

refresh_entity_summaries async ¶

detect_entity_communities¶

detect_entity_communities async ¶

Result Dataclasses¶

ClusteringResult¶

initialize `async` ¶

write `async` ¶

retrieve `async` ¶

get `async` ¶

get_all `async` ¶

delete `async` ¶

delete_all `async` ¶

entities `async` ¶

close `async` ¶

WriteResult `dataclass` ¶

RetrieveResult `dataclass` ¶

FactDetail `dataclass` ¶

EntityDetail `dataclass` ¶

ScoredFact `dataclass` ¶

PipelineTrace `dataclass` ¶

MemoryConfig `dataclass` ¶

complete `async` ¶

embed `async` ¶

embed_one `async` ¶

complete `async` ¶

embed `async` ¶

embed_one `async` ¶

cluster_user_facts `async` ¶

detect_communities `async` ¶

run_consolidation `async` ¶

run_profile_consolidation `async` ¶

run_memify `async` ¶

compute_entity_importance `async` ¶

refresh_entity_summaries `async` ¶

detect_entity_communities `async` ¶

ClusteringResult `dataclass` ¶

CommunityDetectionResult `dataclass` ¶

ConsolidationResult `dataclass` ¶

MemifyResult `dataclass` ¶

EntityImportanceResult `dataclass` ¶

SummaryRefreshResult `dataclass` ¶