Configuration

memharness is fully configurable via Python code, YAML files, or environment variables.

Configuration Methods

1. Python Code

from memharness import MemoryHarness
from memharness.config import (
    Config,
    SummarizationConfig,
    ConsolidationConfig,
    GCConfig,
    EntityExtractionConfig,
    ContextAssemblyConfig,
)

memory = MemoryHarness(
    backend="postgresql://localhost/memharness",
    config=Config(
        summarization=SummarizationConfig(
            enabled=True,
            triggers=[
                {"condition": "age > 7d", "memory_type": "conversational"},
            ],
            keep_originals=True,
            originals_ttl="365d",
        ),
        consolidation=ConsolidationConfig(
            enabled=True,
            schedule="0 3 * * *",
            similarity_threshold=0.9,
        ),
        gc=GCConfig(
            enabled=True,
            schedule="0 4 * * 0",
            archive_after="90d",
            delete_after="365d",
        ),
        entity_extraction=EntityExtractionConfig(
            enabled=True,
            mode="on_write",
        ),
        context_assembly=ContextAssemblyConfig(
            default_max_tokens=4000,
            priorities={
                "conversational": 0.40,
                "knowledge_base": 0.30,
                "entity": 0.15,
                "workflow": 0.10,
                "summary": 0.05,
            },
        ),
    ),
)

2. YAML File

# memharness.yaml
backend: postgresql://localhost/memharness

conversational:
  max_messages_per_thread: 1000
  default_ttl: null

summarization:
  enabled: true
  triggers:
    - condition: "age > 7d"
      memory_type: conversational
    - condition: "message_count > 50"
      memory_type: conversational
  keep_originals: true
  originals_ttl: 365d

consolidation:
  enabled: true
  schedule: "0 3 * * *"
  similarity_threshold: 0.9

gc:
  enabled: true
  schedule: "0 4 * * 0"
  archive_after: 90d
  delete_after: 365d

entity_extraction:
  enabled: true
  mode: on_write
  types:
    - PERSON
    - ORG
    - PLACE
    - CONCEPT

context_assembly:
  default_max_tokens: 4000
  priorities:
    conversational: 0.40
    knowledge_base: 0.30
    entity: 0.15
    workflow: 0.10
    summary: 0.05

memory = MemoryHarness.from_config("memharness.yaml")

3. Environment Variables

# Backend
export MEMHARNESS_BACKEND=postgresql://localhost/memharness

# Quick toggles
export MEMHARNESS_SUMMARIZATION_ENABLED=true
export MEMHARNESS_CONSOLIDATION_ENABLED=true
export MEMHARNESS_GC_ENABLED=true

# LLM for agents
export MEMHARNESS_LLM_PROVIDER=openai
export OPENAI_API_KEY=sk-...

memory = MemoryHarness.from_env()

Configuration Options

Backend

backend: postgresql://user:pass@host:5432/db
# or
backend: sqlite:///path/to/memory.db
# or
backend: memory://

Summarization

Option	Type	Default	Description
enabled	bool	true	Enable summarization
triggers	list	[]	Conditions that trigger summarization
keep_originals	bool	true	Archive originals (don't delete)
originals_ttl	str	"365d"	How long to keep originals

Trigger conditions:

• age > 7d — Messages older than 7 days
• message_count > 50 — More than 50 messages in thread
• context_usage > 80% — Context window above 80%

Consolidation

Option	Type	Default	Description
enabled	bool	true	Enable consolidation
schedule	str	"0 3 * * *"	Cron schedule
similarity_threshold	float	0.9	Merge if > 90% similar
memory_types	list	["entity"]	Types to consolidate

Garbage Collection

Option	Type	Default	Description
enabled	bool	true	Enable GC
schedule	str	"0 4 * * 0"	Cron schedule (weekly)
archive_after	str	"90d"	Move to cold storage
delete_after	str	"365d"	Delete from cold
protect_referenced	bool	true	Don't delete if referenced

Entity Extraction

Option	Type	Default	Description
enabled	bool	true	Enable entity extraction
mode	str	"on_write"	"on_write", "batch", or "disabled"
types	list	[...]	Entity types to extract
min_confidence	float	0.7	Minimum confidence

Context Assembly

Option	Type	Default	Description
default_max_tokens	int	4000	Default token budget
priorities	dict	{...}	Token allocation per type
expand_summaries	bool	false	Auto-expand summaries

Duration Format

Durations can be specified as:

• 7d — 7 days
• 24h — 24 hours
• 30m — 30 minutes
• 365d — 1 year

Cron Format

Standard 5-field cron: minute hour day-of-month month day-of-week

Examples:

• 0 3 * * * — Daily at 3 AM
• 0 4 * * 0 — Weekly on Sunday at 4 AM
• 0 */6 * * * — Every 6 hours

Per-Namespace Overrides

# Global config
memory = MemoryHarness(backend="...", config=global_config)

# Override for specific namespace
memory.configure_namespace(
    namespace=("org", "enterprise"),
    config=Config(
        summarization=SummarizationConfig(
            triggers=[{"condition": "age > 30d"}],  # Keep longer
        ),
        gc=GCConfig(
            delete_after="7y",  # 7 year retention
        ),
    ),
)

Runtime Updates

# Update config without restart
await memory.update_config(
    summarization={"triggers": [{"condition": "age > 14d"}]}
)

# Enable/disable agents
await memory.disable_agent("consolidation")
await memory.enable_agent("consolidation")