> ## Documentation Index
> Fetch the complete documentation index at: https://docs.hydradb.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Recall – Overview

> Quick reference for all Recall endpoints, their differences, and when to call each.

## Which recall endpoint should I use?

```mermaid theme={"dark"}
flowchart LR

%% HydraDB Dark Mode class definitions
classDef standard fill:#0f172a,stroke:#334155,stroke-width:2px,color:#f8fafc;
classDef innovation fill:#CC4515,stroke:#FF571A,stroke-width:3px,color:#ffffff,font-weight:bold;
classDef success fill:#0f172a,stroke:#10b981,stroke-width:2px,color:#f8fafc;
classDef failure fill:#0f172a,stroke:#ef4444,stroke-width:2px,color:#f8fafc;

%% Nodes
Q([I want to retrieve context])
KM([Knowledge or memories?])
KType([What kind of query?])
FR([/recall/full_recall/])
RP([/recall/recall_preferences/])
BR([/recall/boolean_recall/])

%% Links
Q --> KM
KM -- "Documents, files, app sources" --> KType
KM -- "User preferences, conversations" --> RP
KType -- "Semantic question" --> FR
KType -- "Exact term/phrase lookup" --> BR

%% Classes
class Q,KM,KType standard;
class FR innovation;
class RP standard;
class BR standard;

%% Link style
linkStyle default stroke:#64748b,stroke-width:2px;
```

## Endpoint reference

| Endpoint                                                                   | Method | Purpose                          | Returns                                       |
| -------------------------------------------------------------------------- | ------ | -------------------------------- | --------------------------------------------- |
| [`/recall/full_recall`](/api-reference/endpoint/full-recall)               | `POST` | Hybrid recall over knowledge     | Ranked chunks + optional graph context        |
| [`/recall/recall_preferences`](/api-reference/endpoint/recall-preferences) | `POST` | Hybrid recall over user memories | Ranked memory chunks + optional graph context |
| [`/recall/boolean_recall`](/api-reference/endpoint/boolean-recall)         | `POST` | Exact-match full-text search     | BM25-ranked chunks                            |

## Side-by-side comparison

|                | `full_recall`              | `recall_preferences` | `boolean_recall`                  |
| -------------- | -------------------------- | -------------------- | --------------------------------- |
| **Searches**   | Knowledge                  | Memories             | Either (via `search_mode`)        |
| **Match type** | Semantic + graph           | Semantic + graph     | Exact term/phrase                 |
| **Returns**    | Chunks                     | Chunks               | Chunks                            |
| **Latency**    | Medium                     | Medium               | Lowest                            |
| **Best for**   | Document Q\&A, RAG context | Personalization      | Compliance lookups, exact matches |

## Modes

Two of the three endpoints (`full_recall`, `recall_preferences`) accept a `mode` parameter:

| Mode             | Behavior                                                    | When to use                                                              |
| ---------------- | ----------------------------------------------------------- | ------------------------------------------------------------------------ |
| `fast` (default) | Single query pass, no reranking                             | Real-time chat, autocomplete, simple lookups                             |
| `thinking`       | Multi-query expansion, reranking, forceful-relation context | Complex queries, customer-facing answers, anything where quality matters |

`boolean_recall` doesn't have modes – it's deterministic by design.

## Common parameters

`full_recall` and `recall_preferences` share most parameters:

| Parameter          | Default | Purpose                             |
| ------------------ | ------- | ----------------------------------- |
| `query`            | –       | The user's input                    |
| `mode`             | `fast`  | Retrieval quality vs latency        |
| `alpha`            | `0.8`   | Hybrid weight (semantic vs keyword) |
| `recency_bias`     | `0.0`   | Prefer newer content                |
| `metadata_filters` | –       | Deterministic scoping               |
| `graph_context`    | `false` | Include entity relationships        |

## Typical patterns

### Personalized Q\&A

```python theme={"dark"}
# 1. Recall user's preferences first
prefs = client.recall.recall_preferences(
    tenant_id="acme",
    sub_tenant_id="user_alex",
    query="how does the user like answers formatted?",
)

# 2. Then recall the actual knowledge
result = client.recall.full_recall(
    tenant_id="acme",
    query="What's our refund policy?",
    additional_context=format_prefs(prefs),  # use prefs as extra signal
    mode="thinking",
)
```

### Document Q\&A with citations

```python theme={"dark"}
# Retrieve context, then pass to your LLM of choice
result = client.recall.full_recall(
    tenant_id="acme",
    query="What's our refund policy?",
    mode="thinking",
    graph_context=True,
)

# Build prompt for your LLM with retrieved chunks
context = "\n\n".join(c.chunk_content for c in result.chunks)
sources = [c.source_title for c in result.chunks]

# answer = your_llm.complete(f"Context:\n{context}\n\nQuestion: ...")
# cite using `sources`
```

### Compliance lookup

```python theme={"dark"}
# Exact match for a regulatory term
result = client.recall.boolean_recall(
    tenant_id="acme",
    query="GDPR Article 17",
    operator="phrase",
    search_mode="sources",
)
```

## Key concepts

**Hybrid retrieval** – Combines semantic similarity (embeddings) with keyword matching (BM25). The `alpha` parameter controls the weight. Default `0.8` favors semantic.

**Graph context** – Entity relationships extracted from your data. Optional. Helps surface "how things connect" in addition to "what's similar."

**`thinking` mode** – Expands the query into multiple sub-queries, retrieves for each, and reranks the combined results. Higher quality, higher latency.

**Forceful relations** – Sources can be explicitly linked at ingestion. In `thinking` mode, recall surfaces context from these linked sources via the `additional_context` field.

## Related sections

* [Essentials → Recall](/essentials/recall) – conceptual overview, the multi-stage pipeline
* [Essentials → Context Graphs](/essentials/context-graphs) – how graph context is built and used
* [API Reference → Memories](/api-reference/endpoint/add-memory) – ingest memories before recall
* [API Reference → Ingestion](/api-reference/endpoint/upload-knowledge) – ingest knowledge before recall