Cortex Modes

Cortex is the memory consolidation engine. It transforms raw memories into structured, organized knowledge through 8 named processing modes. Each mode runs a pipeline of specialized processors with different goals — from gentle tidying to aggressive cleanup.

How Cortex Works

You trigger a mode — manually via the Cortex tab in the QUI Core dashboard, or via a schedule
A snapshot is taken — the current memory state is saved so you can revert if needed
Processors run in sequence — each mode runs its pipeline of processors, transforming memories step by step
Results persist — consolidated memories, updated associations, and archived originals are saved

Cortex reads from and writes to the same Memory Service database. It uses the character's configured LLM (via Anima) for any processing that requires AI reasoning.

The 8 Modes

Meditation

Gentle consolidation. Decays low-importance memories, resolves duplicates, strengthens associations between related memories, and re-evaluates importance scores. Extracts anxiety patterns and unresolved tensions for insight.

Best for: Regular maintenance. Run weekly to keep memories tidy without aggressive changes.

Psychedelic

Creative association building. Scrambles existing association patterns and creates novel links between distant memories. Merges adjacent clusters and inverts importance (boosting undervalued memories, dampening overvalued ones). Extracts unexpected parallels and hidden metaphors.

Best for: Breaking out of rigid memory patterns. Useful when a character seems stuck in repetitive thinking.

Metalhead

Aggressive cleanup. Prunes weak associations, resolves all duplicates, compresses old memories, applies strong decay, and defragments the database.

Warning: Metalhead is NOT reversible — it permanently deletes memories. Use with caution.

Best for: Characters with bloated memory that need a hard reset to essential knowledge only.

Contemplation

Deep introspection. Runs multiple analysis profiles to extract narrative threads, wisdom, unresolved business, and identity evolution. Synthesizes findings into a reflective narrative.

Best for: Understanding a character's journey. Generates insight documents about patterns, growth, and recurring themes.

Supermind

Comprehensive analysis with large context. Fetches memories based on mode configuration, clusters them by theme, samples representatives, compresses clusters, and creates overview links.

Best for: Building high-level summaries of large memory collections. Good for characters with many memories that need structural organization.

Hyperfocus

Topic-specific consolidation. Requires a topic filter — you specify what topic to focus on (e.g., "database optimization" or "customer feedback"). Fetches only relevant memories, resolves duplicates within the topic, strengthens topic-internal associations, compresses related memories, and extracts deep-dive insights.

Best for: When you want to consolidate knowledge about a specific domain without touching unrelated memories.

Nostalgia

Processes older memories (90+ days). Performs sentiment analysis on old content, boosts positive memories, strengthens positive associations, and generates warm synthesis narratives.

Best for: Characters with long histories. Helps retain and reinforce positive experiences while letting negative ones naturally fade.

Running Cortex

Manual Execution

Open the Cortex tab in the QUI Core dashboard
Select a character
Choose a mode
For Hyperfocus: enter the topic filter
Click Run Now

The execution progress appears in the jobs panel. You can monitor which processors are running and see results as they complete.

Scheduled Execution

Individual processors can be scheduled via cron expressions. Default schedules are configured per processor — for example, decay runs daily at 2 AM, cleanup runs daily at 5 AM, and conversation rollup runs weekly.

Schedules can be customized per character in the Cortex tab or via the Anima Visual Builder's Cortex nodes.

Snapshots and Revert

Before every mode execution, Cortex creates a snapshot of the character's memory state. If a mode produces unexpected results, you can revert to the snapshot from the Cortex tab.

Snapshots auto-expire after 7 days.

Exception: Metalhead mode is NOT reversible — it permanently deletes data rather than archiving it.

Processors Under the Hood

Each mode runs a pipeline of processors. There are 22 processors total, including:

deep_memory — time-based L0→L1 compression
compress — semantic similarity grouping
string_rollup — conversation summarization
pattern_clustering — association graph management
straighten — duplicate and contradiction detection
decay — time-based importance reduction
focus_melt — AI-powered importance re-evaluation
topic_cluster — topic-focused consolidation
cleanup — maintenance and pruning
analysis — insight extraction using 7 built-in profiles
sentiment_analysis — emotional content detection
synthesis — combine analyses into narratives

You don't need to understand individual processors to use modes — modes are pre-configured pipelines designed for specific purposes.

Configuration

Cortex processing is configured per character through the Anima Visual Builder. Add Cortex processor and mode nodes to the character's canvas to:

Enable or disable specific processors
Customize processor settings (thresholds, batch sizes, age limits)
Set up scheduled processing
Choose which modes are available for manual runs

If no Cortex configuration is set on a character, default settings apply (fail-open).

Updated on Mar 21, 2026

Memory