Anchor Engine Settings Reference

June 10, 2026 · View on GitHub

Date: 2026-05-20
Purpose: Consolidated reference for all user_settings.json configuration options

Summary Table

Setting Path	Default Value	Type	Category
Server
`server.host`	`"0.0.0.0"`	string	Server
`server.port`	`3160`	number	Server
`server.api_key`	`""` (empty)	string	Server
GitHub
`github.token`	`"ghp_..."` (placeholder)	string	GitHub
`github.default_branch`	`"main"`	string	GitHub
Paths
`paths.anchor_root`	`<ANCHOR_ROOT>` (placeholder)	string	Paths
(derived)	Auto-generated from root	-	Paths
Encryption
`encryption.enabled`	`false`	boolean	Encryption
`encryption.auto_encrypt_on_ingest`	`true`	boolean	Encryption
`encryption.detect_nsfw`	`false`	boolean	Encryption
Search
`search.strategy`	`"hybrid"`	string ('hybrid' \| 'semantic' \| 'exact')	Search
`search.max_chars_default`	`5000`	number	Search
`search.fts_window_size`	`1500`	number	Search
Resource Management
`resource_management.gc_cooldown_ms`	`30000` (30s)	number	Resource Mgmt
`resource_management.max_atoms_in_memory`	`2000`	number	Resource Mgmt
`resource_management.monitoring_interval_ms`	`30000` (30s)	number	Resource Mgmt
Memory Management
`memory.heap_pressure_mb`	`500` (env: ANCHOR_HEAP_PRESSURE_MB)	number	Memory
`memory.throttle_start_mb`	`800` (env: ANCHOR_THROTTLE_START_MB)	number	Memory
`memory.throttle_max_mb`	`1200` (env: ANCHOR_THROTTLE_MAX_MB)	number	Memory
`memory.emergency_stop_mb`	`1500` (env: ANCHOR_EMERGENCY_STOP_MB)	number	Memory
`memory.enable_streaming_results`	`true`	boolean	Memory
Watcher
`watcher.debounce_ms`	`2000` (unused/legacy)	number	Watcher
`watcher.stability_threshold_ms`	`2000` (2s)	number	Watcher
`watcher.extra_paths`	`[]` (empty array)	string[]	Watcher
`watcher.auto_start`	`false`	boolean	Watcher
Context & Inference
`context.relevance_weight`	`0.7`	number	Context
`context.recency_weight`	`0.3`	number	Context
Limits
`limits.max_file_size_bytes`	`104857600` (~100MB)	number	Limits
`limits.max_content_length_chars`	`5000`	number	Limits
Database
`database.wipe_on_startup`	`true`	boolean	Database
`database.shared_buffers_mb`	`256`	number	Database
`database.effective_cache_size_mb`	`512`	number	Database
`database.work_mem_mb`	`32` (code) / `16` (template)	number	Database
`database.maintenance_work_mem_mb`	`32`	number	Database
Low Resource Mode
`low_resource.enabled`	`false`	boolean	Low Resource
Ingestion
`ingestion.concept_density`	`"medium"` (env: ANCHOR_CONCEPT_DENSITY)	string ('low' \| 'medium' \| 'high')	Ingestion
`ingestion.tag_threshold`	`0.7` (env: ANCHOR_TAG_THRESHOLD)	number (0-1)	Ingestion
`ingestion.dedup_strength`	`"medium"` (env: ANCHOR_DEDUP_STRENGTH)	string ('light' \| 'medium' \| 'aggressive')	Ingestion
`ingestion.token_budget_default`	`2000` (env: ANCHOR_TOKEN_BUDGET_DEFAULT)	number	Ingestion
MCP
`mcp.enabled`	`true`	boolean	MCP
`mcp.require_api_key`	`false`	boolean	MCP
Agent Memory
`agent_memory.auto_distill`	`true`	boolean	Agent Memory
`agent_memory.checkpoint_interval_messages`	`50`	number	Agent Memory
Logging
`logging.level`	`"info"`	string ('debug' \| 'info' \| 'warn' \| 'error')	Logging
`logging.structured`	`true`	boolean	Logging

Server Configuration

Network and server binding settings.

`server.host`

Default: "0.0.0.0"
Description: Network interface the server binds to
When to change: Use "127.0.0.1" for local-only development; use "0.0.0.0" for production with proper firewall rules

`server.port`

Default: 3160
Description: Port number the server listens on
Why this default: Uncommon port reduces collision risk, in typical API range (80-9000)

`server.api_key`

Default: "" (empty - optional)
Description: Optional API key for authentication
Security implication: Empty = no auth required. Enable for production with sensitive data

GitHub Authentication

Settings for cloning and interacting with GitHub repositories.

`github.token`

Default: "ghp_YOUR_PERSONAL_ACCESS_TOKEN_HERE" (placeholder)
Description: GitHub Personal Access Token for cloning private repos and higher rate limits
Security implication: NEVER commit to git! Store in user_settings.json (gitignored) or use environment variable
Required scopes: 'repo' for private repos, 'public_repo' for public-only access

`github.default_branch`

Default: "main"
Description: Default branch to checkout when cloning repos
Why this default: "main" is modern standard (replaced "master")

Path Configuration

All paths are relative to anchor_root unless specified as absolute.

`paths.anchor_root`

Default: <ANCHOR_ROOT> (placeholder)
Description: Base directory for all Anchor Engine data
Recommendation: Use a dedicated directory like ~/anchored_data or ~/.anchor

Derived Paths

The following paths are automatically derived from anchor_root:

paths.notebook, paths.inbox, paths.external_inbox, and others

Encryption Settings

End-to-end encryption configuration for sensitive content.

`encryption.enabled`

Default: false
Description: Enable end-to-end encryption for sensitive content
Security implication: When enabled, all content is encrypted at rest and in transit

`encryption.auto_encrypt_on_ingest`

Default: true
Description: Automatically encrypt content during ingestion
Why this default: Security-first approach - encrypt by default

`encryption.detect_nsfw`

Default: false
Description: Enable NSFW (Not Safe For Work) content detection
Security implication: Disabled to avoid false positives and privacy concerns

Search Configuration

Search algorithm and query processing settings.

`search.strategy`

Default: "hybrid"
Type: string ('hybrid' | 'semantic' | 'exact')
Description: Selects search algorithm strategy to use
Options:
- "hybrid": Uses both semantic and exact matching (default)
- "semantic": Vector-based search only
- "exact": FTS-only precise lookups

`search.max_chars_default`

Default: 5000 (~1.25k tokens)
Description: Maximum characters to process per search query by default
Why this default: Mobile-friendly design - lower limit for bandwidth-constrained users

`search.fts_window_size`

Default: 1500 tokens (~375 words)
Description: Window size for full-text search token overlap analysis
Impact: Larger windows = more context but slower search

Resource Management

Garbage collection and memory pressure monitoring settings.

`resource_management.gc_cooldown_ms`

Default: 30000 (30 seconds)
Description: Cooldown period between forced garbage collection cycles
Impact: Prevents excessive GC that could slow down operations

`resource_management.max_atoms_in_memory`

Default: 2000
Description: Maximum atoms to keep in memory at once before eviction
Tradeoff: Higher values = more memory usage but better search performance

`resource_management.monitoring_interval_ms`

Default: 30000 (30 seconds)
Description: Interval for resource monitoring checks
Impact: Balances responsiveness with CPU overhead

Memory Management

Heap pressure thresholds and streaming settings. Many can be overridden via environment variables.

`memory.heap_pressure_mb`

Default: 500 (env: ANCHOR_HEAP_PRESSURE_MB)
Description: Heap memory level that triggers "high pressure" state
Impact: Triggers aggressive memory management when heap usage exceeds this threshold

`memory.throttle_start_mb`

Default: 800 (env: ANCHOR_THROTTLE_START_MB)
Description: Heap memory level that triggers throttling of non-critical operations
Behavior: Below this = normal operation; above = start limiting features

`memory.throttle_max_mb`

Default: 1200 (env: ANCHOR_THROTTLE_MAX_MB)
Description: Maximum heap memory before aggressive throttling kicks in
Behavior: Between throttle_start and throttle_max = moderate throttling; above = aggressive

`memory.emergency_stop_mb`

Default: 1500 (env: ANCHOR_EMERGENCY_STOP_MB)
Description: Critical heap memory level that triggers emergency stop measures
Impact: Last line of defense before OOM errors - stops all non-critical operations

`memory.enable_streaming_results`

Default: true
Description: Enable streaming search results instead of loading all at once
Benefit: Reduces memory usage for large result sets

Watcher (File Watching)

Automated file change detection and ingestion settings.

`watcher.debounce_ms`

Default: 2000 ms (from template, unused/legacy in code)
Description: Debounce delay for file change detection
Note: Not directly used - actual debouncing uses separate constant INGESTION_DEBOUNCE_MS = 30000

`watcher.stability_threshold_ms`

Default: 2000 ms (2 seconds)
Description: Chokidar's awaitWriteFinish.stabilityThreshold - waits this many ms after file write completes before triggering ingestion event
Tradeoff: Lower = faster response but risk of incomplete files; Higher = more reliable but slower

`watcher.extra_paths`

Default: [] (empty array)
Description: Additional directories to watch beyond default inbox/external-inbox
Usage: Set in user_settings.json: "watcher": { "extra_paths": ["./my-custom-dir"] }
Endpoints: Dynamically managed via /v1/watchdog/add-path and /v1/watchdog/remove-path

`watcher.auto_start`

Default: false
Description: Controls whether watchdog starts automatically on server startup
Security consideration: User wants this OFF by default for safety - prevents automatic ingestion when server starts with potentially bad/empty paths
Enable via: Environment variable AUTO_START_WATCHDOG=true OR settings file

Context & Inference

Relevance scoring weights for context selection.

`context.relevance_weight`

Default: 0.7
Description: Weight given to relevance signals in context scoring

`context.recency_weight`

Default: 0.3
Description: Weight given to recency signals in context scoring
Tradeoff: Lower weight means older but relevant content still considered valuable

Limits & Thresholds

Input validation and resource limits.

`limits.max_file_size_bytes`

Default: 104857600 (~100MB)
Description: Maximum file size accepted during ingestion
Purpose: Prevents extremely large files from consuming excessive resources

`limits.max_content_length_chars`

Default: 5000
Description: Maximum content length for API requests
Purpose: Reasonable limit to prevent abuse and resource exhaustion

Database Configuration

PostgreSQL performance tuning settings. Note: Comments in code incorrectly reference SQLite - these are PostgreSQL configs.

`database.wipe_on_startup`

Default: true
Description: Controls whether database is wiped and rebuilt on server startup
Security consideration: true means fresh database each start - prevents data persistence across restarts
Note: Setting to false retains existing database but index may become stale if schema changed

`database.shared_buffers_mb`

Default: 256 MB
Description: PostgreSQL shared buffer cache size - memory for database caching
Performance tuning: Higher values improve query performance but use more RAM
Fallback: Falls back to 64MB if not configured

`database.effective_cache_size_mb`

Default: 512 MB
Description: PostgreSQL effective cache size hint - helps query planner optimize queries
Recommendation: Should be set to ~50% of available RAM for best results
Note: This is a hint only - doesn't actually allocate memory

`database.work_mem_mb`

Default: 32 MB (code) / 16 MB (template mismatch)
Description: PostgreSQL work memory for sorts and hashes - larger = faster complex queries
Tradeoff: Larger values speed up sorts/hashes but use more memory per operation
Risk: OOM errors if set too high for available RAM

`database.maintenance_work_mem_mb` (BONUS DISCOVERY)

Default: 32 MB
Description: PostgreSQL maintenance operations memory (VACUUM, CREATE INDEX)
Note: Not in original template - added for completeness

Low Resource Mode

Optimized settings for systems with limited resources.

`low_resource.enabled`

Default: false
Description: Enable optimized mode for systems with limited resources
When enabled:
- Lower recall thresholds (5-20 results instead of higher)
- Reduced concurrency (1 worker)
- Smaller batch sizes for ingestion and search

Ingestion Settings

Content processing and concept extraction settings. Many can be overridden via environment variables.

`ingestion.concept_density`

Default: "medium" (env: ANCHOR_CONCEPT_DENSITY)
Type: string ('low' | 'medium' | 'high')
Description: Controls how densely concepts are extracted from content
Options:
- "high": More concepts extracted (better recall, higher token usage)
- "low": Fewer concepts (faster ingestion, lower recall)

`ingestion.tag_threshold`

Default: 0.7 (env: ANCHOR_TAG_THRESHOLD)
Type: number (0-1)
Description: Minimum similarity score for a token to be considered a meaningful concept
Impact: Higher threshold = fewer, more precise tags; Lower = more tags but potentially noisy

`ingestion.dedup_strength`

Default: "medium" (env: ANCHOR_DEDUP_STRENGTH)
Type: string ('light' | 'medium' | 'aggressive')
Description: Controls aggressiveness of duplicate content detection during ingestion
Options:
- "aggressive": More duplicates removed (cleaner index, slower ingestion)
- "light": Fewer duplicates removed (faster ingestion, potentially redundant content)

`ingestion.token_budget_default`

Default: 2000 (env: ANCHOR_TOKEN_BUDGET_DEFAULT)
Description: Default token budget for content processing during ingestion
Impact: Higher budget = more content per request but higher cost/latency

MCP (Model Context Protocol)

Settings for the Model Context Protocol server.

`mcp.enabled`

Default: true
Description: Enable Model Context Protocol server for AI agent memory
When to disable: If not using advanced AI agent features

`mcp.require_api_key`

Default: false
Description: Require API key to access MCP endpoints
Security implication: False = open access; enable for production security

Agent Memory

Conversation distillation and memory management settings.

`agent_memory.auto_distill`

Default: true
Description: Automatically distill agent conversations periodically
Benefit: Maintains conversation history without manual intervention

`agent_memory.checkpoint_interval_messages`

Default: 50 messages
Description: Number of messages between distillation checkpoints
Tradeoff: Balances memory usage with conversation fidelity

Logging Configuration

Log level and format settings.

`logging.level`

Default: "info"
Type: string ('debug' | 'info' | 'warn' | 'error')
Description: Log level for output verbosity

`logging.structured`

Default: true
Description: Enable structured JSON logging format
Benefit: Structured logs are easier to parse and analyze programmatically

01-watcher-settings.md - Detailed watcher configuration exploration
02-database-settings.md - Database performance tuning details
03-resource-memory-settings.md - Resource limits and memory management
04-search-ingestion-settings.md - Search algorithms and ingestion quality
05-user-settings-map.md - Complete settings map with reasoning

Security Recommendations

Critical Settings to Configure for Production

github.token - Never commit to git, use environment variable or local settings file
server.api_key - Enable for production with sensitive data
mcp.require_api_key - Enable for MCP endpoints in production
encryption.enabled - Consider enabling for sensitive content

Common Pitfalls to Avoid

Don't set database.wipe_on_startup=false without understanding implications (index may become stale)
Don't enable encryption.enabled without proper password management
Don't commit user_settings.json to git - it's in .gitignore
Be cautious with watcher.auto_start=true - prevents automatic ingestion when server starts with bad paths

Environment Variable Overrides

Many settings can be overridden via environment variables for deployment flexibility:

Setting	Env Var	Description
`memory.heap_pressure_mb`	`ANCHOR_HEAP_PRESSURE_MB`	Heap pressure threshold
`memory.throttle_start_mb`	`ANCHOR_THROTTLE_START_MB`	Throttling start point
`memory.throttle_max_mb`	`ANCHOR_THROTTLE_MAX_MB`	Aggressive throttling point
`memory.emergency_stop_mb`	`ANCHOR_EMERGENCY_STOP_MB`	Emergency stop threshold
`search.strategy`	N/A	Search algorithm selection
`ingestion.concept_density`	`ANCHOR_CONCEPT_DENSITY`	Concept extraction density
`ingestion.tag_threshold`	`ANCHOR_TAG_THRESHOLD`	Tag similarity threshold
`ingestion.dedup_strength`	`ANCHOR_DEDUP_STRENGTH`	Duplicate detection strength
`ingestion.token_budget_default`	`ANCHOR_TOKEN_BUDGET_DEFAULT`	Token budget for ingestion
`watcher.auto_start`	`AUTO_START_WATCHDOG`	Auto-start watchdog service

Generated from consolidated settings exploration reports. Last updated: 2026-05-20