Advanced Memory System

Multi-tiered memory with quality scoring and graph support

Multi-tiered memory with short-term, long-term, entity, and user-specific storage — enhanced by quality scoring and automatic extraction.

from praisonaiagents import Agent, MemoryConfig

agent = Agent(
    name="memory-agent",
    instructions="Remember context across turns.",
    memory=MemoryConfig(use_long_term=True),
)
agent.start("What did we decide about the launch date?")

The user chats across turns; multi-tier memory stores and scores context for later retrieval.

How It Works

Key Features

Separate short-term and long-term memory systems

4-metric quality assessment for stored memories

User, agent, and run-specific memory scoping

Automatic entity extraction and storage

Optional Neo4j/Memgraph for relationships

Quality-based filtering and relevance ranking

Quick Start

1. Simple — enable with True

   from praisonaiagents import Agent
   
   agent = Agent(
       name="Assistant",
       instructions="You are a helpful assistant with memory.",
       memory=True
   )
   
   agent.start("Remember that I prefer Python for data science")
   
2. With config

   from praisonaiagents import Agent
   
   agent = Agent(
       name="Assistant",
       instructions="You are a helpful assistant with memory.",
       memory={
           "backend": "rag",
           "user_id": "user123",
           "use_embedding": True
       }
   )
   
   

---

Direct Memory Access

from praisonaiagents import Memory

# Initialise memory
memory = Memory(
    config={
        "provider": "rag",
        "use_embedding": True,
        "short_db": ".praison/short_term.db",
        "long_db": ".praison/long_term.db"
    }
)

# Store memories with quality
memory.store_long_term(
    text="The user prefers technical explanations",
    memory={"user_id": "user123"},
    metadata={"type": "preference"},
    quality=0.9
)

# Search memories — scope with the first-class user_id kwarg
results = memory.search_long_term(
    query="user preferences",
    user_id="user123",
    min_quality=0.7
)

from praisonaiagents import Memory

memory = Memory(config={"provider": "rag"})

memory.store_long_term(
    text="Quantum computing uses qubits for computation",
    completeness=0.95,
    relevance=0.90,
    clarity=0.88,
    accuracy=0.92,
    memory={"user_id": "user123"}
)

Memory Tiers

Short-term Memory (STM)

Short-term memory is cleared between sessions and used for immediate context.

# Store temporary context and capture the returned id
mid = memory.store_short_term(
    text="User is asking about Python programming",
    metadata={"user_id": "user123", "context": "current_conversation"}
)

# Search recent context
context = memory.search_short_term(
    query="Python",
    limit=5
)

# Targeted delete uses the id (returns True on success)
memory.delete_short_term(mid)

# Or wipe everything
memory.reset_short_term()

memory.reset_short_term() / .reset_long_term() now work on the default rag / chroma provider, not just sqlite / in_memory and dakera. On sqlite / in_memory providers, reset_* / delete_* delegate to the active adapter (mirroring search_short_term), so they operate on the adapter-created short_term_memory / long_term_memory tables.

On the default rag provider, resetting short-term also clears long-term — Chroma stores both tiers in one collection. Use DakeraMemoryAdapter if you need tier isolation.

Adapters that don’t implement reset_short_term / reset_long_term now log a warning stating the tier was NOT cleared, instead of silently succeeding — check logs after upgrade.

Since PraisonAI PR #4020, Chroma-backed memory (the default) supports reset. On earlier releases, Memory().reset_short_term() / .reset_long_term() was a documented, callable, exception-free API that quietly did nothing.

Long-term Memory (LTM)

Long-term memory persists across sessions and stores important information.

# Store persistent knowledge
memory.store_long_term(
    text="User works as a data scientist at TechCorp",
    memory={"user_id": "user123"},
    quality=0.95,
    metadata={"type": "user_info", "category": "professional"}
)

# Retrieve with quality filter — user_id is a first-class kwarg
memories = memory.search_long_term(
    query="user profession",
    user_id="user123",
    min_quality=0.8
)

Metadata types (ChromaDB / RAG backend)

ChromaDB only stores primitives — PraisonAI sanitises your metadata dict before writing so Chroma writes never fail.

Value type	What gets stored
str, int, float, bool	passed through unchanged
None	key dropped entirely (does not round-trip)
dict, list, tuple, any other type	coerced via str(value) — stored as its repr-style string

from praisonaiagents import Memory

memory = Memory(config={"provider": "rag"})

memory.store_long_term(
    text="Project atlas kicked off",
    metadata={
        "project": "atlas",              # stored as "atlas"
        "priority": 3,                    # stored as 3
        "shipped": False,                 # stored as False
        "owner": None,                    # DROPPED — not stored
        "tags": ["backend", "urgent"],    # stored as the string "['backend', 'urgent']"
        "config": {"env": "prod"},        # stored as the string "{'env': 'prod'}"
    },
)

ChromaDB itself rejects non-primitive metadata values; PraisonAI coerces them to strings so writes never fail. If you need to filter or round-trip complex metadata, keep it as JSON in a str field (e.g. metadata={"tags_json": json.dumps(tags)}) so it survives untouched. The mem0, mongodb, and dakera adapters do not apply this coercion — they accept richer metadata natively.

Entity Memory

Entity memory stores information about specific people, places, or things.

# Store entity information
memory.store_entity(
    name="TechCorp",
    text="TechCorp is a leading AI company founded in 2020",
    entity_type="organization",
    metadata={"industry": "technology", "size": "large"}
)

# Search entity information
entity_info = memory.search_entity(
    name="TechCorp",
    query="company details"
)

User Memory

User memory stores personalised information and preferences.

# Store user preferences
memory.store_user_memory(
    memory={"user_id": "user123"},
    text="Prefers concise explanations with code examples",
    extra={"preference_type": "communication_style"}
)

# Retrieve user context
user_context = memory.search_user_memory(
    memory={"user_id": "user123"},
    query="preferences",
    limit=3
)

Configuration Options

MemoryConfig SDK Reference

Full parameter reference for MemoryConfig

MemoryBackend options:

Backend	Value	Description
File	"file"	JSON files (default, zero dependencies)
SQLite	"sqlite"	SQLite database
In-memory	"in_memory"	Lightweight in-process store (dev/testing)
Chroma	"chroma"	ChromaDB vector store
Mem0	"mem0"	Mem0 cloud service (requires mem0ai)
MongoDB	"mongodb"	MongoDB
Dakera	"dakera"	Self-hosted decay-weighted memory server (requires praisonaiagents[dakera])

redis, valkey, and postgres are not memory backends — passing them raises ValueError. See Unimplemented Backends Raise below. For Redis/Postgres state persistence use db(state_url=...) / db(database_url=...).

MemoryConfig parameters:

Option	Type	Default	Description
backend	str	"file"	Storage backend (see above)
user_id	str | None	None	Scope memory to a specific user
session_id	str | None	None	Scope memory to a specific session
auto_memory	bool	False	Auto-extract and store key facts
claude_memory	bool	False	Use Claude’s native memory (Anthropic models)
config	dict | None	None	Provider-specific configuration
learn	bool | LearnConfig | None	None	Enable continuous learning
history	bool	False	Auto-inject session history into context
history_limit	int	10	Max messages to inject from history
auto_save	str | None	None	Auto-save session with this name

RAG Configuration (Default)

memory_config = {
    "provider": "rag",
    "use_embedding": True,
    "short_db": ".praison/short_term.db",
    "long_db": ".praison/long_term.db",
    "entity_db": ".praison/entity.db",
    "rag_db_path": ".praison/chroma_db",
    "embedding_model": "text-embedding-3-small"
}

Mem0 Configuration

Requires the optional memory extras:

pip install "praisonaiagents[memory]"

If the package is missing, configuring "provider": "mem0" raises:

ImportError: mem0ai is not installed. Run: pip install 'praisonaiagents[memory]'

memory_config = {
    "provider": "mem0",
    "config": {
        "api_key": "your-mem0-api-key",
        "org_id": "your-org-id",
        "project_id": "your-project-id"
    }
}

Dakera Configuration

Requires the optional Dakera extra:

pip install "praisonaiagents[dakera]"

If the package is missing, configuring "provider": "dakera" raises:

ImportError: dakera is required for the dakera adapter. Install with: pip install 'praisonaiagents[dakera]' (or: pip install dakera)

memory_config = {
    "provider": "dakera",
    "config": {
        "url": "http://localhost:3000",      # or DAKERA_URL / DAKERA_API_URL env var
        "api_key": "dk-...",                 # or DAKERA_API_KEY env var
        "agent_id": "my-agent",             # or DAKERA_AGENT_ID env var
        "short_term_type": "working",
        "long_term_type": "episodic",
        "default_importance": 0.5,
    }
}

See Dakera Memory for the full guide including self-hosted deployment, tier mapping, and delete/reset operations.

Graph Memory Configuration

Requires the optional memory extras:

pip install "praisonaiagents[memory]"

memory_config = {
    "provider": "mem0",
    "config": {
        "graph_store": {
            "provider": "neo4j",
            "config": {
                "url": "bolt://localhost:7687",
                "username": "neo4j",
                "password": "password"
            }
        },
        "vector_store": {
            "provider": "qdrant",
            "config": {
                "host": "localhost",
                "port": 6333
            }
        },
        "version": "v1.1"
    }
}

Unimplemented Backends Raise

Requesting a backend that has no registered adapter raises a ValueError — there is no silent fallback to a local file store.

redis, valkey, and postgres are not memory adapters. Earlier releases accepted them but silently stored to a local file, so data never reached Redis/Postgres. Requesting one now raises:

ValueError: Memory backend 'redis' is not implemented - no redis memory adapter
is registered. Valid backends: chroma, dakera, file, in_memory, mem0, mongodb,
sqlite. To keep using redis, pass a live store (memory=<instance> or db(url=...)),
or register an adapter with register_memory_adapter('redis', MyAdapter).

Use Redis persistence (db(state_url=...)) instead.

Configured backend	Behavior	Store/search path
file	native (default)	FileMemory (direct)
sqlite	native adapter	memory_adapter.* (direct)
chroma	ChromaDB adapter (aliases chromadb, rag)	memory_adapter.*
mem0	Mem0 client	direct Mem0 client
mongodb	MongoDB adapter	memory_adapter.*
dakera	Dakera adapter	memory_adapter.*
in_memory	in-process adapter (alias none)	memory_adapter.*
redis / postgres / valkey	raises ValueError	use db(state_url=...) (Redis) or db(database_url=...) (Postgres)

All registered adapters route store_*, search_*, reset_*, and delete_* through memory_adapter.*. See Custom Memory Adapters.

A registered adapter whose optional dependency is missing logs a logger.warning and the memory layer degrades gracefully — this is distinct from requesting an unimplemented backend, which raises ValueError.

Keep using Redis or Postgres

Pass a live store via db(state_url=...) for Redis (or db(database_url=...) for Postgres):

from praisonaiagents import Agent, MemoryConfig, db

agent = Agent(
    name="Assistant",
    instructions="You remember user preferences.",
    memory=MemoryConfig(db=db(state_url="redis://localhost:6379/0"), user_id="user_123"),
)
agent.start("Remember I prefer dark mode")

Register a custom adapter (escape hatch)

Register an adapter for the name you want, then backend="redis" is accepted:

from praisonaiagents.memory.adapters import register_memory_adapter

register_memory_adapter("redis", MyRedisAdapter)
agent = Agent(memory=MemoryConfig(backend="redis"))  # now accepted

---

Quality Scoring System

Quality Metrics

Completeness

Measures how complete and comprehensive the information is

Relevance

Measures how relevant the information is to the context

Clarity

Measures how clear and understandable the information is

Accuracy

Measures the factual accuracy of the information

Quality Calculation

# Default weights
weights = {
    "completeness": 0.25,
    "relevance": 0.35,
    "clarity": 0.20,
    "accuracy": 0.20
}

# Overall quality = weighted average
quality = (
    completeness * weights["completeness"] +
    relevance * weights["relevance"] +
    clarity * weights["clarity"] +
    accuracy * weights["accuracy"]
)

Advanced Features

Scoped search with user_id and metadata_filter

search_short_term and search_long_term accept user_id and metadata_filter as first-class kwargs to scope results per tenant.

from praisonaiagents import Agent, Memory

memory = Memory(config={"provider": "rag", "use_embedding": True})
agent = Agent(
    name="TenantAssistant",
    instructions="Answer scoped to the current tenant.",
    memory=True,
)

# Multi-tenant memory scoping
results = memory.search_long_term(
    query="project details",
    user_id="tenant_a",             # tenant isolation — always wins
    metadata_filter={"project": "atlas"},  # extra scoping
    limit=10,
)

user_id is now enforced twice on every search — once at the Memory layer (existing) and again inside SearchMixin.search_short_term / SearchMixin.search_long_term as of PraisonAI PR #4819. The lower layer auto-derives the same user_id metadata filter, so adapters that accept user_id but ignore it in their own query (Chroma/RAG, SQLite, in-memory) still isolate per-tenant results. If a caller passes both an explicit user_id and a conflicting metadata_filter["user_id"], the explicit user_id always wins — a caller cannot widen scope to another tenant’s data.

user_id takes precedence over any metadata_filter["user_id"], so a route can never widen scope to another tenant. Applied across every backend — Mem0, MongoDB (with and without vector search), ChromaDB, memory-adapter, and the SQLite fallback. When a metadata_filter is present, each backend over-fetches by 10× so results ranked beyond limit still survive the post-filter, then truncates back to limit.

Direct SearchMixin use (subclassing or embedding in a custom store) previously depended on the adapter to filter by user_id. If you have such code, no change is required — the filter is applied automatically now — but you may safely remove any duplicate post-filtering you added as a workaround.

Before PR #3855, user_id and metadata_filter were silently absorbed into **kwargs and never applied — per-user scoping was broken on the class-body path. Pass them explicitly, not inside a memory={...} dict.

Cache-Optimised Context

Memory results are deterministically ordered for prompt caching effectiveness. Use build_context_for_task() with explicit output control for manual prompt assembly.

from praisonaiagents import Memory

memory = Memory(config={"provider": "rag"})

context = memory.build_context_for_task(
    task_descr="Research the latest AI developments",
    user_id="user_123",
    max_items=3,
    include_in_output=True  # Force include for prompt assembly
)

# Construct cache-friendly prompt structure
system_prompt = f"System instructions and tools\n\n{context}\n\nUser message: {{user_input}}"

See Prompt Caching for the full guide.

Context Building

Build comprehensive context for tasks:

# Automatically merge relevant memories
context = memory.build_context_for_task(
    task_descr="Write a Python tutorial",
    memory={"user_id": "user123"},
    additional="Focus on data science applications",
    max_items=5
)

# Context includes:
# - Relevant long-term memories
# - User preferences
# - Recent short-term context

Task Output Finalisation

Store task results with quality assessment:

from praisonaiagents import TaskOutput

# Create task output
output = TaskOutput(
    raw="Tutorial content...",
    agent_name="Writer",
    task_description="Write Python tutorial"
)

# Finalise with quality scoring
memory.finalize_task_output(
    task_output=output,
    memory={"user_id": "user123"}
)

Memory Citations

Automatically cite memory sources:

# Memories include source information
result = memory.search_long_term("Python", user_id="user123")
# Returns: {
#     'text': 'Python is...',
#     'metadata': {'source': 'conversation_2024_01_15', ...}
# }

Memory Reranking

Enhance search results with intelligent reranking based on relevance scores:

from praisonaiagents import Agent, AgentTeam

# Create agent
agent = Agent(
    name="Researcher",
    role="Research assistant"
)

# Initialize with memory enabled
agents = AgentTeam(
    agents=[agent],
    tasks=[],
    memory=True
)

# Search with reranking for better relevance
if agents.shared_memory:
    results = agents.shared_memory.search_long_term(
        "quantum computing applications",
        limit=10,
        rerank=True,  # Enable reranking
        relevance_cutoff=0.7  # Minimum relevance score
    )

Reranking Features

1. Semantic Reranking: Re-scores results based on semantic similarity
2. Context-Aware Ranking: Considers current context when ranking
3. Quality-Weighted Ranking: Combines relevance with quality scores

# Advanced reranking with multiple criteria
results = agents.shared_memory.search_long_term(
    query="machine learning frameworks",
    limit=20,  # Retrieve more initially
    rerank=True,  # Enable reranking
    relevance_cutoff=0.6,  # Minimum relevance
    rerank_config={
        "method": "semantic",  # or "hybrid", "quality-weighted"
        "top_k": 5,  # Return top 5 after reranking
        "consider_recency": True,  # Factor in timestamp
        "boost_user_memories": True  # Prioritize user-specific memories
    }
)

Custom Reranking Logic

Implement custom reranking strategies:

from datetime import datetime
from praisonaiagents import RerankStrategy

class DomainSpecificReranker(RerankStrategy):
    def rerank(self, results, query, context=None):
        """Custom reranking based on domain knowledge"""
        scored_results = []
        
        for result in results:
            score = result['relevance']
            
            # Boost technical content
            if 'technical' in result.get('metadata', {}).get('tags', []):
                score *= 1.2
                
            # Boost recent memories
            if result.get('timestamp'):
                days_old = (datetime.now() - result['timestamp']).days
                recency_boost = 1.0 + (0.1 * max(0, 30 - days_old) / 30)
                score *= recency_boost
                
            scored_results.append((score, result))
        
        # Sort by score and return top results
        scored_results.sort(key=lambda x: x[0], reverse=True)
        return [result for score, result in scored_results]

# Use custom reranker
# Note: Assuming 'agents' is defined from previous examples
if agents.shared_memory:
    agents.shared_memory.set_rerank_strategy(DomainSpecificReranker())

Reranking Performance

Optimize reranking for large result sets:

# Efficient two-stage retrieval and reranking
# Stage 1: Fast retrieval of candidates
candidates = agents.shared_memory.search_long_term(
    query="data analysis techniques",
    limit=50,  # Get more candidates
    rerank=False  # Skip reranking in first stage
)

# Stage 2: Precise reranking of top candidates
if len(candidates) > 10:
    reranked = agents.shared_memory.rerank_results(
        results=candidates[:20],  # Rerank top 20
        query="data analysis techniques",
        method="hybrid",
        relevance_cutoff=0.8
    )
else:
    reranked = candidates

Performance Optimisation

Batch Operations

# Batch store memories
memories = [
    ("Fact 1", 0.9),
    ("Fact 2", 0.85),
    ("Fact 3", 0.95)
]

for text, quality in memories:
    memory.store_long_term(
        text=text,
        quality=quality,
        memory={"user_id": "user123"}
    )

Complete Example

Personal Assistant with Memory

from praisonaiagents import Agent, Task, AgentTeam
from praisonaiagents import Memory

# Configure memory with quality scoring
memory_config = {
    "provider": "rag",
    "use_embedding": True,
    "embedding_model": "text-embedding-3-small"
}

# Create memory instance
memory = Memory(memory_config)

# Create assistant agent
assistant = Agent(
    name="Personal Assistant",
    instructions="""You are a personal assistant with perfect memory.
    Remember user preferences, important information, and context.
    Always use your memory to provide personalised responses.""",
    memory={"user_id": "john_doe"}  # Memory with user isolation
)

# Create task with memory integration
task = Task(
    description="Help me plan my day based on my preferences",
    agent=assistant,
    expected_output="Personalised daily schedule"
)

# Run the system
agents = AgentTeam(
    agents=[assistant],
    tasks=[task],
    memory=True  # Enable shared memory
)

# Memory automatically:
# 1. Retrieves user preferences
# 2. Searches relevant past interactions
# 3. Builds context for the task
# 4. Stores the interaction with quality score
result = agents.start()

# Store user feedback as high-quality memory
memory.store_long_term(
    text="User preferred morning schedule with exercise first",
    memory={"user_id": "john_doe"},
    completeness=0.9,
    relevance=1.0,
    clarity=0.95,
    accuracy=1.0
)

Best Practices

Set quality thresholds for search

Use min_quality=0.8 for critical information retrieval. Use lower thresholds (0.5–0.7) for exploratory searches where recall is more important.

Scope memories per user and agent

Always pass user_id for user-specific memories, agent_id for agent-scoped shared knowledge, and run_id for ephemeral session context.

Enable embeddings for semantic search

Set use_embedding=True in the memory config for semantic (not just keyword) retrieval. Use text-embedding-3-small for cost efficiency.

Use AutoMemory for automatic extraction

Enable MemoryConfig(auto_memory=True) to automatically extract preferences, facts, and entities from conversations without manual store_* calls.

AutoMemory

AutoMemory automatically extracts structured information from conversations using configurable patterns — no manual store_* calls needed.

How It Works

AutoMemory wraps FileMemory with a pattern-based extraction engine. After each agent interaction, it scans the conversation for matching patterns (preferences, facts, dates, etc.) and stores them automatically.

Quick Start

from praisonaiagents import Agent, MemoryConfig

# Enable auto-memory extraction
agent = Agent(
    name="Assistant",
    instructions="You are a helpful assistant.",
    memory=MemoryConfig(auto_memory=True, user_id="user123")
)

# Patterns are extracted automatically from conversation
agent.start("I prefer Python for data science and I'm based in London")
# AutoMemory stores: preference="Python for data science", location="London"

Custom Patterns

from praisonaiagents import AutoMemory

# Create with custom patterns
auto_mem = AutoMemory(
    user_id="user123",
    patterns={
        "preference": r"(?:I prefer|I like|I love)\s+(.+)",
        "location": r"(?:I'm based in|I live in|I'm from)\s+(.+)",
        "name": r"(?:My name is|I'm called|call me)\s+(\w+)",
        "fact": r"(?:I work at|I study at|my job is)\s+(.+)",
    }
)

# Search extracted memories
results = auto_mem.search("user preferences")

MemoryConfig Integration

Use MemoryConfig.auto_memory to enable AutoMemory as part of the consolidated memory parameter:

from praisonaiagents import Agent, MemoryConfig

agent = Agent(
    instructions="You are a helpful assistant.",
    memory=MemoryConfig(
        backend="file",
        auto_memory=True,   # Enable pattern-based extraction
        user_id="user123",
        session_id="session456",
    )
)

Parameter	Type	Default	Description
auto_memory	bool	False	Enable automatic memory extraction
backend	str	"file"	Storage backend (file, sqlite, chroma, mongodb, mem0, dakera, in_memory)
user_id	str	None	User ID for scoping memories

---

Related

Knowledge

Integrate with document knowledge bases

Prompt Caching

Optimise memory context for prompt caching

Vector Store

Store and query embeddings with a pluggable, namespace-aware backend