Skip to content

Latest commit

 

History

History
117 lines (92 loc) · 3.64 KB

Configuration.md

File metadata and controls

117 lines (92 loc) · 3.64 KB

tinyMem Configuration Reference

This guide lists all available configuration options for .tinyMem/config.toml.

Core Structure

A standard configuration file looks like this:

[proxy]
port = 8080
base_url = "http://localhost:11434/v1"

[llm]
model = "llama3"
timeout = 120

[recall]
max_items = 10

[cove]
enabled = true
confidence_threshold = 0.6

[logging]
level = "info"
file = "tinymem.log"

Section Details

[proxy]

Settings for the HTTP proxy server.

  • port (integer): The port tinyMem listens on. Default: 8080.
  • base_url (string): The upstream LLM provider's base URL (e.g., http://localhost:11434/v1).

[llm]

Settings for the internal LLM client (used for CoVe, summarization, and proxying).

  • model (string): The model identifier to send to the backend.
  • timeout (integer): Request timeout in seconds. Default: 120.
  • api_key_env (string): Name of the environment variable containing the API key.
  • base_url (string): Override backend URL specifically for internal tasks (defaults to [proxy].base_url).

[recall]

Settings for memory retrieval.

  • max_items (integer): Maximum number of memories to retrieve per query. Default: 10.
  • max_tokens (integer): Maximum total tokens for injected context. Default: 2000.

[cove]

Chain-of-Verification settings for truth validation.

  • enabled (boolean): Enable CoVe. Default: false.
  • confidence_threshold (float): Minimum confidence (0.0 - 1.0) to accept a fact. Default: 0.6.
  • max_candidates (integer): Number of candidate facts to generate/check. Default: 3.

[logging]

  • level (string): Log verbosity. Options: debug, info, warn, error. Default: info.
  • file (string): Log file path relative to .tinyMem/. Default: tinymem.log.

[database]

Database maintenance and retention policy settings.

  • auto_maintenance (boolean): Enable automatic database maintenance (PRAGMA optimize + incremental_vacuum). Default: true.
  • maintenance_interval_hours (integer): How often to run maintenance (in hours). Default: 24 (daily).

Example:

[database]
auto_maintenance = true
maintenance_interval_hours = 24  # Run daily

[database.retention]
max_age_days = 90  # Delete memories older than 90 days
max_count = 10000  # Keep only 10,000 most recent memories
exclude_types = ["fact", "decision"]  # Never delete facts or decisions

[database.retention]

Retention policy for automatic memory cleanup.

  • max_age_days (integer): Delete memories older than this many days. Default: 0 (unlimited).
  • max_count (integer): Keep only the N most recent memories. Default: 0 (unlimited).
  • exclude_types (array of strings): Memory types to never delete. Default: ["fact"].

Retention Policy Behavior:

  • Age-based and count-based retention can be used together
  • Both policies respect the exclude_types list
  • Retention runs on startup and periodically (if auto_maintenance is enabled)
  • Use max_age_days = 0 and max_count = 0 for unlimited retention
  • Excluded types are always preserved regardless of age or count

Example Use Cases:

Long-term project (keep everything, just maintain):

[database]
auto_maintenance = true
maintenance_interval_hours = 168  # Weekly

Short-lived sessions (aggressive cleanup):

[database]
auto_maintenance = true
maintenance_interval_hours = 1  # Hourly

[database.retention]
max_age_days = 7  # One week
exclude_types = ["fact", "decision"]

Size-limited (keep only recent memories):

[database]
auto_maintenance = true

[database.retention]
max_count = 5000  # Keep 5,000 most recent
exclude_types = ["fact"]