fleet-memory/hindsight-docs/docs/developer/api/memory-banks.md

sidebar_position
6

Memory bank Identity

Configure memory bank personality, background, and behavior.

import Tabs from '@theme/Tabs'; import TabItem from '@theme/TabItem';

Creating an Memory bank

from hindsight_client import Hindsight

client = Hindsight(base_url="http://localhost:8888")

client.create_agent(
    agent_id="my-agent",
    name="Research Assistant",
    background="I am a research assistant specializing in machine learning",
    personality={
        "openness": 0.8,
        "conscientiousness": 0.7,
        "extraversion": 0.5,
        "agreeableness": 0.6,
        "neuroticism": 0.3,
        "bias_strength": 0.5
    }
)

import { OpenAPI, ManagementService } from '@hindsight/client';

OpenAPI.BASE = 'http://localhost:8888';

await ManagementService.createAgentApiAgentsAgentIdPut('my-agent', {
    name: 'Research Assistant',
    background: 'I am a research assistant specializing in machine learning',
    personality: {
        openness: 0.8,
        conscientiousness: 0.7,
        extraversion: 0.5,
        agreeableness: 0.6,
        neuroticism: 0.3,
        bias_strength: 0.5
    }
});

# Set background
hindsight agent background my-agent "I am a research assistant specializing in ML"

# Set personality
hindsight memory bank personality my-agent \
    --openness 0.8 \
    --conscientiousness 0.7 \
    --extraversion 0.5 \
    --agreeableness 0.6 \
    --neuroticism 0.3 \
    --bias-strength 0.5

Personality Traits (Big Five)

Each trait is scored 0.0 to 1.0:

Trait	Low (0.0)	High (1.0)
Openness	Conventional, prefers proven methods	Curious, embraces new ideas
Conscientiousness	Flexible, spontaneous	Organized, systematic
Extraversion	Reserved, independent	Outgoing, collaborative
Agreeableness	Direct, analytical	Cooperative, diplomatic
Neuroticism	Calm, optimistic	Risk-aware, cautious

How Traits Affect Behavior

Openness influences how the memory bank weighs new vs. established ideas:

# High openness agent
"Let's try this new framework—it looks promising!"

# Low openness agent
"Let's stick with the proven solution we know works."

Conscientiousness affects structure and thoroughness:

# High conscientiousness agent
"Here's a detailed, step-by-step analysis..."

# Low conscientiousness agent
"Quick take: this should work, let's try it."

Extraversion shapes collaboration preferences:

# High extraversion agent
"We should get the team together to discuss this."

# Low extraversion agent
"I'll analyze this independently and share my findings."

Agreeableness affects how disagreements are handled:

# High agreeableness agent
"That's a valid point. Perhaps we can find a middle ground..."

# Low agreeableness agent
"Actually, the data doesn't support that conclusion."

Neuroticism influences risk assessment:

# High neuroticism agent
"We should consider what could go wrong here..."

# Low neuroticism agent
"The risks seem manageable, let's proceed."

Background

The background is a first-person narrative providing agent context:

client.create_agent(
    agent_id="financial-advisor",
    background="""I am a conservative financial advisor with 20 years of experience.
    I prioritize capital preservation over aggressive growth.
    I have seen multiple market crashes and believe in diversification."""
)

Background influences:

• How questions are interpreted
• Perspective in responses
• Opinion formation context

Merging Background

New background information is merged intelligently:

# Original background
client.create_agent(
    agent_id="assistant",
    background="I am a helpful AI assistant"
)

# Add more context (merged, not replaced)
client.update_background(
    agent_id="assistant",
    background="I specialize in Python programming"
)

# Result: "I am a helpful AI assistant. I specialize in Python programming."

Merging rules:

• Conflicts: New overwrites old
• Additions: Non-conflicting info is added
• Normalization: "You are..." → "I am..."

Getting Memory bank Profile

profile = client.get_profile(agent_id="my-agent")

print(f"Background: {profile['background']}")
print(f"Personality: {profile['personality']}")

hindsight memory bank profile my-agent

Updating Personality

client.update_personality(
    agent_id="my-agent",
    openness=0.9,
    conscientiousness=0.8
)

Listing Memory banks

memory banks = client.list_agents()
for agent in memory banks:
    print(agent["agent_id"])

hindsight agent list

Default Values

If not specified, memory banks use neutral defaults:

{
    "openness": 0.5,
    "conscientiousness": 0.5,
    "extraversion": 0.5,
    "agreeableness": 0.5,
    "neuroticism": 0.5,
    "bias_strength": 0.5,
    "background": ""
}

Personality Templates

Common personality configurations:

Use Case	O	C	E	A	N	Bias
Customer Support	0.5	0.7	0.6	0.9	0.3	0.4
Code Reviewer	0.4	0.9	0.3	0.4	0.5	0.6
Creative Writer	0.9	0.4	0.7	0.6	0.5	0.7
Risk Analyst	0.3	0.9	0.3	0.4	0.8	0.6
Research Assistant	0.8	0.8	0.4	0.5	0.4	0.5
Neutral (default)	0.5	0.5	0.5	0.5	0.5	0.5

# Customer support agent
client.create_agent(
    agent_id="support",
    background="I am a friendly customer support agent",
    personality={
        "openness": 0.5,
        "conscientiousness": 0.7,
        "extraversion": 0.6,
        "agreeableness": 0.9,  # Very diplomatic
        "neuroticism": 0.3,    # Calm under pressure
        "bias_strength": 0.4
    }
)

# Code reviewer agent
client.create_agent(
    agent_id="reviewer",
    background="I am a thorough code reviewer focused on quality",
    personality={
        "openness": 0.4,       # Prefers proven patterns
        "conscientiousness": 0.9,  # Very thorough
        "extraversion": 0.3,
        "agreeableness": 0.4,  # Direct feedback
        "neuroticism": 0.5,
        "bias_strength": 0.6
    }
)

Memory bank Isolation

Each agent has:

• Separate memories — memory banks don't share memories
• Own personality — traits are per-agent
• Independent opinions — formed from their own experiences

# Store to agent A
client.store(agent_id="agent-a", content="Python is great")

# Memory bank B doesn't see it
results = client.search(agent_id="agent-b", query="Python")
# Returns empty