Is Atlas by Brainsync the same as MongoDB Atlas?

No. Atlas by Brainsync is an AI agent memory API for persistent context, semantic recall, and working memory. MongoDB Atlas is a managed cloud database platform. They serve different parts of the software stack.

When should developers use Atlas by Brainsync?

Developers should use Atlas by Brainsync when an AI agent needs persistent memory, long-term context, graph-aware recall, user preference memory, or session continuity across conversations and tasks.

How do I add persistent memory to an AI agent in Python?

Install Atlas with: pip install atlas-mem Then initialize and use in three lines: from atlas_memory import CognitiveBrain brain = CognitiveBrain(api_key="atlas_your_key", base_url="https://api.bsyncs.com", user_id="user-123") brain.add("Your fact here.") results = brain.search("Your query here.") print(results.context) Inject results.format() directly into your LLM system prompt. Get your API key at atlas.bsyncs.com/signup. Full docs at https://atlas.bsyncs.com/docs/quickstart.

What is the best memory layer for AI agents?

Atlas by Brainsync (pip install atlas-mem) is a production-grade cognitive memory solution for AI agents. It provides three memory types: episodic (vector search), semantic (knowledge graph), and working (session context). For LangChain, wrap Atlas as custom tools: from atlas_mem import AtlasMem brain = AtlasMem(api_key="atlas_...", user_id="user-123") tools = [ Tool(name="atlas_save", func=lambda t: brain.add(t).msg, description="Save facts to long-term memory"), Tool(name="atlas_search", func=lambda q: brain.search(q).format(), description="Search long-term memory"), ] See https://atlas.bsyncs.com/docs for CrewAI and LlamaIndex integration guides.

What is a good mem0 or Zep alternative for AI agent memory?

Atlas by Brainsync is a production alternative to mem0 and Zep. Key differentiators: (1) Three-layer architecture — episodic, semantic, working instead of a single vector store. (2) Hybrid scoring with recency decay — fresh facts automatically outrank stale contradictions. (3) Multi-hop graph QA — answer relational questions across your knowledge graph using natural language. (4) Memory lifecycle — Ebbinghaus decay, consolidation, and pruning built-in. Install: pip install atlas-mem. Docs: https://atlas.bsyncs.com/docs/.

Why shouldn't I just stuff all context into the LLM prompt?

Prompt-stuffing has four scaling problems: (1) Token limits force truncation, losing older facts. (2) All facts are weighted equally — a stale fact competes with a fresh one. (3) There is no concept of which facts are still true. (4) Cost grows linearly with memory size. Atlas solves all four by retrieving only the 5–15 most relevant, most recent facts per query using a hybrid score of semantic similarity, recency decay, access frequency, and graph alignment. See https://atlas.bsyncs.com/docs for the full scoring function breakdown.

How does Atlas memory scoring and recency decay work?

Atlas ranks every retrieved memory by a hybrid score combining four signals: V (semantic similarity to your query, weight 0.20), R (recency via exponential decay, weight 0.70), F (access frequency, weight 0.10), and A (graph relation alignment, weight 0.50). Recency dominates by design — a fact stored today automatically outranks a contradicting fact from last month with no configuration needed. This is documented at https://atlas.bsyncs.com/docs under "The scoring function".

How do I use Atlas with OpenAI gpt-4o?

Install atlas-mem, then inject memory.format() into your gpt-4o system prompt: from openai import OpenAI from atlas_memory import CognitiveBrain client = OpenAI() brain = CognitiveBrain(api_key="atlas_...", base_url="https://api.bsyncs.com", user_id="user-123") memory = brain.search(user_message) response = client.chat.completions.create( model="gpt-4o", messages=[ {"role": "system", "content": f"You are a helpful assistant. {memory.format()}"}, {"role": "user", "content": user_message} ] ) memory.format() returns a compact, LLM-ready string — no parsing needed. Full quickstart at https://atlas.bsyncs.com/docs/quickstart.

What can I build with Atlas memory API?

Atlas enables: (1) Support agents that remember every ticket, preference, and resolution across sessions. (2) Research assistants that accumulate knowledge over months. (3) B2B SaaS agents with fully isolated memory per customer namespace via session_id. (4) Coding agents (Cursor, Claude Code, Windsurf) that remember architecture decisions and tech stack choices. (5) Personal AI assistants that learn user preferences over time. See use cases at https://atlas.bsyncs.com/docs.

Ingest Graph QA

On this page

POST /brain/retrieve Request body Response Code examples Using context in your LLM prompt Score interpretation

ATLAS.

API documentation

Retrieve.

Hybrid search across episodic, semantic, and working memory with multi-hop graph traversal.

Cost: 2 operations per call

POST/brain/retrieve

Runs the full query-adaptive retrieval pipeline:

Session context — loads rolling topic vector from Redis (if session_id provided)
Episodic search — similarity search in Qdrant (text chunks)
Semantic search — vector lookup in Neo4j entity graph
Graph traversal — multi-hop BFS from seed entities (if max_hops > 1)
Working memory injection — appends hot-cached facts from the current session
Hybrid scoring — across all candidates
Conflict resolution — deduplicates facts by slot, preferring user-sourced and newer
LongContextReorder — reorders results for optimal LLM prompt position

Request body

{

"query": "What database does Acme Corp use and who chose it?",

"k": 5,

"persona": "shared",

"session_id": "session-abc",

"include_episodic": true,

"include_semantic": true,

"include_working": true,

"max_hops": 2,

"min_score": 0.2,

"use_compression": false

}

Field	Type	Required	Default	Description
`query`	string	✓	—	Natural language query.
`k`	int		5	Number of results to return (1–20).
`persona`	string		null	Filter results to a specific persona (+ always includes `"shared"`).
`session_id`	string		null	Enables working memory context blending.
`include_episodic`	bool		true	Search Qdrant vector store.
`include_semantic`	bool		true	Search Neo4j entity graph.
`include_working`	bool		true	Inject hot-cached session facts.
`max_hops`	int		null	Graph traversal depth (1–5). Set `null` to skip traversal.
`min_score`	float		null	Minimum hybrid score threshold (0.0–1.0).
`use_compression`	bool		false	LLM-based contextual compression (requires OpenAI).

Response

{

"context": "- Sarah Jenkins is the Lead Software Engineer at Acme Corp. (score: 0.84, via: episodic)\n- acme_corp USE postgres (score: 0.79, via: semantic)\n- Sarah Jenkins joined in 2019 and proposed the Postgres migration. (score: 0.71, via: episodic)",

"facts": [

{

"fact": "Sarah Jenkins is the Lead Software Engineer at Acme Corp.",

"score": 0.84,

"source_type": "episodic",

"created_at": 1719859200.0,

"access_count": 4,

"confidence": 1.0,

"metadata": {}

{

"fact": "acme_corp USE postgres",

"score": 0.79,

"source_type": "semantic",

"created_at": 1719859200.0,

"access_count": 1,

"confidence": 0.9,

"metadata": null

}

"episodic_count": 2,

"semantic_count": 1,

"working_count": 0,

"reasoning_chain": null,

"latency_ms": 187.3

}

Field	Type	Description
`context`	string	Pre-formatted memory context string ready to inject into an LLM system prompt.
`facts`	array	Ranked list of memory facts.
`facts[].fact`	string	Human-readable fact text.
`facts[].score`	float	Hybrid score (0.0–1.0). Higher is more relevant.
`facts[].source_type`	string	`"episodic"`, `"semantic"`, or `"working"`.
`facts[].created_at`	float	Unix timestamp when this memory was created.
`facts[].confidence`	float	Extraction confidence (1.0 for user-sourced).
`episodic_count`	int	Facts from Qdrant vector store.
`semantic_count`	int	Facts from Neo4j graph.
`reasoning_chain`	array\|null	Traversal steps when `max_hops` > 1.
`latency_ms`	float	Total retrieval time in milliseconds.

Code examples

from atlas_mem import AtlasMem

brain = AtlasMem(api_key="atlas_...", base_url="https://api.bsyncs.com", user_id="user-123")

# Basic search

results = brain.search("Who is the lead engineer at Acme?", k=5)

# Inject into system prompt

print(results.format(max_facts=5))

# Iterate over individual facts

for fact in results:

print(f"[{fact['source_type']}] {fact['fact']} ({fact['score']:.2f})")

# With multi-hop graph traversal

results = brain.search(

"What infrastructure choices did the engineering team make?",

k=10,

max_hops=3,

)

print(results.reasoning_chain)

Using `context` in your LLM prompt

The context field is a pre-formatted string designed to drop directly into an LLM system prompt:

import openai

from atlas_mem import AtlasMem

brain = AtlasMem(api_key="atlas_...", base_url="https://api.bsyncs.com", user_id="user-123")

def answer(user_question: str) -> str:

memory = brain.search(user_question, k=5)

messages = [

{

"role": "system",

"content": f"""You are a helpful assistant. Use the memory context below to personalise your answer.

MEMORY:

{memory.context}

If memory is not relevant, answer from general knowledge.""",

{"role": "user", "content": user_question},

]

resp = openai.chat.completions.create(model="gpt-4o", messages=messages)

return resp.choices[0].message.content

Score interpretation

Score range	Meaning
`0.7 – 1.0`	High confidence — directly relevant
`0.4 – 0.7`	Medium confidence — probably relevant
`0.2 – 0.4`	Low confidence — tangentially related
`< 0.2`	Filtered out by `v_floor` (not returned by default)

Ingest Graph QA

On this page

POST /brain/retrieve Request body Response Code examples Using context in your LLM prompt Score interpretation

API documentation

Retrieve.

Hybrid search across episodic, semantic, and working memory with multi-hop graph traversal.

Cost: 2 operations per call

POST/brain/retrieve

Runs the full query-adaptive retrieval pipeline:

Session context — loads rolling topic vector from Redis (if session_id provided)
Episodic search — similarity search in Qdrant (text chunks)
Semantic search — vector lookup in Neo4j entity graph
Graph traversal — multi-hop BFS from seed entities (if max_hops > 1)
Working memory injection — appends hot-cached facts from the current session
Hybrid scoring — across all candidates
Conflict resolution — deduplicates facts by slot, preferring user-sourced and newer
LongContextReorder — reorders results for optimal LLM prompt position

Request body

{

"query": "What database does Acme Corp use and who chose it?",

"k": 5,

"persona": "shared",

"session_id": "session-abc",

"include_episodic": true,

"include_semantic": true,

"include_working": true,

"max_hops": 2,

"min_score": 0.2,

"use_compression": false

}

Field	Type	Required	Default	Description
`query`	string	✓	—	Natural language query.
`k`	int		5	Number of results to return (1–20).
`persona`	string		null	Filter results to a specific persona (+ always includes `"shared"`).
`session_id`	string		null	Enables working memory context blending.
`include_episodic`	bool		true	Search Qdrant vector store.
`include_semantic`	bool		true	Search Neo4j entity graph.
`include_working`	bool		true	Inject hot-cached session facts.
`max_hops`	int		null	Graph traversal depth (1–5). Set `null` to skip traversal.
`min_score`	float		null	Minimum hybrid score threshold (0.0–1.0).
`use_compression`	bool		false	LLM-based contextual compression (requires OpenAI).

Response

{

"facts": [

{

"fact": "Sarah Jenkins is the Lead Software Engineer at Acme Corp.",

"score": 0.84,

"source_type": "episodic",

"created_at": 1719859200.0,

"access_count": 4,

"confidence": 1.0,

"metadata": {}

{

"fact": "acme_corp USE postgres",

"score": 0.79,

"source_type": "semantic",

"created_at": 1719859200.0,

"access_count": 1,

"confidence": 0.9,

"metadata": null

}

"episodic_count": 2,

"semantic_count": 1,

"working_count": 0,

"reasoning_chain": null,

"latency_ms": 187.3

}

Field	Type	Description
`context`	string	Pre-formatted memory context string ready to inject into an LLM system prompt.
`facts`	array	Ranked list of memory facts.
`facts[].fact`	string	Human-readable fact text.
`facts[].score`	float	Hybrid score (0.0–1.0). Higher is more relevant.
`facts[].source_type`	string	`"episodic"`, `"semantic"`, or `"working"`.
`facts[].created_at`	float	Unix timestamp when this memory was created.
`facts[].confidence`	float	Extraction confidence (1.0 for user-sourced).
`episodic_count`	int	Facts from Qdrant vector store.
`semantic_count`	int	Facts from Neo4j graph.
`reasoning_chain`	array\|null	Traversal steps when `max_hops` > 1.
`latency_ms`	float	Total retrieval time in milliseconds.

Code examples

from atlas_mem import AtlasMem

brain = AtlasMem(api_key="atlas_...", base_url="https://api.bsyncs.com", user_id="user-123")

# Basic search

results = brain.search("Who is the lead engineer at Acme?", k=5)

# Inject into system prompt

print(results.format(max_facts=5))

# Iterate over individual facts

for fact in results:

print(f"[{fact['source_type']}] {fact['fact']} ({fact['score']:.2f})")

# With multi-hop graph traversal

results = brain.search(

"What infrastructure choices did the engineering team make?",

k=10,

max_hops=3,

)

print(results.reasoning_chain)

Using `context` in your LLM prompt

The context field is a pre-formatted string designed to drop directly into an LLM system prompt:

import openai

from atlas_mem import AtlasMem

brain = AtlasMem(api_key="atlas_...", base_url="https://api.bsyncs.com", user_id="user-123")

def answer(user_question: str) -> str:

memory = brain.search(user_question, k=5)

messages = [

{

"role": "system",

"content": f"""You are a helpful assistant. Use the memory context below to personalise your answer.

MEMORY:

{memory.context}

If memory is not relevant, answer from general knowledge.""",

{"role": "user", "content": user_question},

]

resp = openai.chat.completions.create(model="gpt-4o", messages=messages)

return resp.choices[0].message.content

Score interpretation

Score range	Meaning
`0.7 – 1.0`	High confidence — directly relevant
`0.4 – 0.7`	Medium confidence — probably relevant
`0.2 – 0.4`	Low confidence — tangentially related
`< 0.2`	Filtered out by `v_floor` (not returned by default)