aim-settings

Memory Settings - Configuration Viewer

Display the current configuration of the AI Memory Module, including collections, types, thresholds, token budgets, and service endpoints.

Usage

# Show all memory settings
/aim-settings

# Show specific section
/aim-settings --section collections
/aim-settings --section types
/aim-settings --section thresholds
/aim-settings --section services
/aim-settings --section agents

Configuration Sections

Collections

Shows the 3 Memory System V2.0 collections:

• code-patterns - Project-specific implementation patterns
• conventions - Cross-project shared conventions
• discussions - Decision context and session summaries

Memory Types (14 total)

Organized by collection:

• code-patterns: implementation, error_fix, refactor, file_pattern
• conventions: rule, guideline, port, naming, structure
• discussions: decision, session, blocker, preference, context

Thresholds

• similarity_threshold - Minimum relevance score for search results (default: 0.7)
• dedup_threshold - Similarity threshold for duplicate detection (default: 0.95)
• max_retrievals - Maximum memories per search (default: 5)
• token_budget - Maximum tokens for context injection (default: 4000, per BP-039)

Services

Shows connection details for:

• Qdrant - Vector database (default: localhost:26350)
• Embedding Service - Jina AI embeddings (default: localhost:28080)
• Monitoring API - Health checks and metrics (default: localhost:28000)
• Streamlit Dashboard - Web UI (default: localhost:28501)
• Grafana - Metrics visualization (default: localhost:23000)
• Prometheus - Metrics storage (default: localhost:29090)
• Pushgateway - Metrics push gateway (default: localhost:29091)

Agent Token Budgets

Shows token allocation per BMAD agent:

• architect: 1500 tokens
• analyst: 1200 tokens
• pm: 1200 tokens
• developer/dev: 1200 tokens
• solo-dev: 1500 tokens
• quick-flow-solo-dev: 1500 tokens
• ux-designer: 1000 tokens
• qa: 1000 tokens
• tea: 1000 tokens
• code-review/code-reviewer: 1200 tokens
• scrum-master/sm: 800 tokens
• tech-writer: 800 tokens
• default: 1000 tokens

Logging

• log_level - Logging verbosity (default: INFO)
• log_format - Log output format (json or text, default: json)

Collection Size Limits

• Warning threshold - 10,000 points
• Critical threshold - 50,000 points

Examples

# View complete configuration
/aim-settings

# Check current thresholds
/aim-settings --section thresholds

# View service endpoints
/aim-settings --section services

# Check agent token budgets
/aim-settings --section agents

# View all memory types
/aim-settings --section types

Python Configuration Reference

Configuration is managed by src/memory/config.py:

from src.memory.config import get_config, AGENT_TOKEN_BUDGETS, get_agent_token_budget

# Get configuration singleton
config = get_config()

# Access settings
print(f"Qdrant: {config.qdrant_host}:{config.qdrant_port}")
print(f"Similarity threshold: {config.similarity_threshold}")
print(f"Max retrievals: {config.max_retrievals}")

# Get agent-specific token budget
budget = get_agent_token_budget("architect")  # Returns 1500

Environment Variables

Configuration can be customized via environment variables or .env file:

# Core thresholds
SIMILARITY_THRESHOLD=0.7    # Retrieval relevance cutoff
DEDUP_THRESHOLD=0.95        # Duplicate detection sensitivity
MAX_RETRIEVALS=5            # Results per search
TOKEN_BUDGET=4000           # Context injection limit (per BP-039)

# Service endpoints
QDRANT_HOST=localhost
QDRANT_PORT=26350
EMBEDDING_HOST=localhost
EMBEDDING_PORT=28080
MONITORING_HOST=localhost
MONITORING_PORT=28000

# Logging
LOG_LEVEL=INFO              # DEBUG, INFO, WARNING, ERROR, CRITICAL
LOG_FORMAT=json             # json or text

# Collection size limits
COLLECTION_SIZE_WARNING=10000
COLLECTION_SIZE_CRITICAL=50000

Configuration Precedence

Settings are loaded in this order (highest priority first):

1. Environment variables
2. .env file in project root
3. Default values

Output Format

The skill displays configuration in organized sections with:

• Current values
• Default values (if different)
• Validation ranges (for thresholds)
• Service URLs with ports
• Memory type mappings to collections

Technical Details

• Type Safety: Uses pydantic-settings v2.6+ for validation
• Immutable: Configuration is frozen (thread-safe)
• Singleton: Single config instance per process (lru_cache)
• Validation: All thresholds validated on load

Related Skills

• /aim-search - Use these settings for memory search
• /aim-status - Check system health and statistics

Notes

• Configuration is loaded once at startup
• Changes to .env require service restart
• All ports use 2XXXX prefix to avoid conflicts
• Token budgets optimized per agent role (architects need more context than scrum masters)