### Dreamer Initialization Output

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Example output from the setup script, showing the creation of directories and the LanceDB 'memories' table.

```text
# Dreamer home: /home/user/.dreamer
#   Created: /home/user/.dreamer/episodes
#   Created: /home/user/.dreamer/episodes/archive
#   Created: /home/user/.dreamer/lancedb
#   Created: /home/user/.dreamer/dream-log
#   Created: /home/user/.dreamer/memory-archive
#   Created: /home/user/.dreamer/workspace
#   Created: /home/user/.dreamer/workspace/docs
#   Created LanceDB 'memories' table (1536-dim vectors)
```

--------------------------------

### Initialize Dreamer and Create Example Episode

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Initialize the Dreamer data directory and create an example episode file for testing purposes.

```bash
python setup.py --example
```

--------------------------------

### Setup and Initialization: setup.py

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Initializes the Dreamer data directory structure and creates the LanceDB memories table. Optionally generates an example episode file.

```APIDOC
## Setup and Initialization: setup.py

### Description
Initializes the Dreamer data directory structure and creates the LanceDB memories table with the required schema for vector storage. Optionally generates an example episode file for testing.

### Usage
```bash
# Initialize with default path (~/.dreamer)
python setup.py

# Custom data directory
python setup.py --home /path/to/data

# Initialize and create example episode
python setup.py --example
```

### Output
```
# Dreamer home: /home/user/.dreamer
#   Created: /home/user/.dreamer/episodes
#   Created: /home/user/.dreamer/episodes/archive
#   Created: /home/user/.dreamer/lancedb
#   Created: /home/user/.dreamer/dream-log
#   Created: /home/user/.dreamer/memory-archive
#   Created: /home/user/.dreamer/workspace
#   Created: /home/user/.dreamer/workspace/docs
#   Created LanceDB 'memories' table (1536-dim vectors)
```
```

--------------------------------

### Systemd Timer Setup for Dreamer

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Example systemd configuration for running Dreamer as a nightly cron job at 3 AM. This includes service and timer unit files, along with commands to enable and start the timer.

```bash
# /etc/systemd/system/dreamer.service
[Unit]
Description=Dreamer memory consolidation
After=network.target

[Service]
Type=oneshot
User=youruser
Environment="DREAMER_HOME=/home/youruser/.dreamer"
Environment="OPENAI_API_KEY=sk-..."
WorkingDirectory=/path/to/dreamer
ExecStart=/usr/bin/python3 dreamer.py --verbose
StandardOutput=append:/var/log/dreamer.log
StandardError=append:/var/log/dreamer.log

# /etc/systemd/system/dreamer.timer
[Unit]
Description=Run Dreamer nightly at 3 AM

[Timer]
OnCalendar=*-*-* 03:00:00
Persistent=true

[Install]
WantedBy=timers.target

# Enable and start
sudo systemctl enable --now dreamer.timer
sudo systemctl status dreamer.timer
```

--------------------------------

### Install Dependencies and Run Dreamer

Source: https://github.com/eesiz/clawdreamer/blob/master/README.md

Install project dependencies, set up environment variables, initialize the data directory and LanceDB table, and then run the main Dreamer script. Ensure your OpenAI API key is configured in the .env file.

```bash
# 1. Install dependencies
pip install -r requirements.txt

# 2. Set up environment
cp .env.example .env
# Edit .env with your OpenAI API key

# 3. Initialize data directory + LanceDB table
python setup.py --example

# 4. Run
python dreamer.py --verbose
```

--------------------------------

### Enable and Start session-flush Systemd Timer

Source: https://github.com/eesiz/clawdreamer/blob/master/README.md

Enables and starts the systemd timer for the `session-flush` script, which automatically generates episode files at 2 AM daily to prevent data loss.

```bash
sudo systemctl enable --now session-flush.timer
```

--------------------------------

### Standalone Dreamer Usage Example

Source: https://github.com/eesiz/clawdreamer/blob/master/README.md

Demonstrates how to use Dreamer standalone by specifying episode files and pointing to a LanceDB store via the DREAMER_HOME environment variable. Run `python setup.py` to initialize the LanceDB table.

```bash
$DREAMER_HOME/episodes/2024-03-15.md
$DREAMER_HOME/episodes/2024-03-16.md
```

--------------------------------

### Dreamer Configuration Parameters

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Configure Dreamer using environment variables. Defaults are defined in config.py. This example shows settings for data directory, embedding providers, LLM providers, and tunable parameters.

```bash
# .env file example

# Data directory
DREAMER_HOME=~/.dreamer

# Embedding provider: openai (default), ollama, sentence-transformers
DREAMER_EMBEDDING_PROVIDER=openai
OPENAI_API_KEY=sk-...
DREAMER_EMBEDDING_DIM=1536

# Ollama settings (if using ollama)
OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_EMBEDDING_MODEL=nomic-embed-text

# LLM provider: openai (default), ollama, minimax
DREAMER_LLM_PROVIDER=openai
OLLAMA_LLM_MODEL=qwen2.5:3b

# Tunable parameters (in config.py)
# CLUSTER_SIMILARITY=0.75      # Threshold for grouping chunks
# DEDUP_SIMILARITY=0.90        # Skip if existing memory is this similar
# CONTRADICTION_SIMILARITY=0.70 # Conflict detection threshold
# IMPORTANCE_DECAY_RATE=0.05   # Daily decay rate
# SOFT_DELETE_THRESHOLD=0.15   # Below this = memory deleted
# MAX_EPISODES_PER_RUN=7       # Max days processed per cycle
# MAX_NEW_MEMORIES=20          # Cap on new memories per cycle
```

--------------------------------

### Example Episode File Content

Source: https://github.com/eesiz/clawdreamer/blob/master/README.md

Markdown content for an episode file, including session notes, user interactions, and API details. This format is used for daily AI agent experiences.

```markdown
# Session Notes - 2024-03-15

## User asked about deployment
Discussed Docker setup. User prefers docker-compose over raw Docker commands.
Decided to use nginx as reverse proxy.

## API Integration
Connected to the payment API. Key endpoint: POST /v1/charges
Rate limit: 100 req/min. Auth via Bearer token.
```

--------------------------------

### Episode File Format Example

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Episodes are markdown files representing daily AI agent experiences. Dreamer processes files named YYYY-MM-DD.md or YYYY-MM-DD-slug.md from the episodes directory.

```markdown
# Session Notes - 2024-03-15

## Project Setup
Discussed deployment strategy. Decided on Docker Compose with nginx reverse proxy.
Database: PostgreSQL 16 with pgvector extension for embeddings.

## API Integration
Connected to the payment gateway API.
- Endpoint: POST /v1/charges
- Rate limit: 100 req/min
- Auth: Bearer token in header
- Webhook URL configured for payment confirmations.

## User Preferences
User prefers dark mode UI. Font size: 14px.
Keyboard shortcut for save: Ctrl+S (not Cmd+S).
```

--------------------------------

### Schedule Daily Cron Job

Source: https://github.com/eesiz/clawdreamer/blob/master/README.md

Example cron job entry to run the dreamer script daily at 3 AM. Ensure the path and log redirection are correctly configured for your environment.

```bash
# Example: run daily at 3 AM
0 3 * * * cd /path/to/dreamer && python3 dreamer.py --verbose >> dream-log/cron.log 2>&1
```

--------------------------------

### Dreamer Cycle JSON Output

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Example JSON output structure detailing the results of the NREM, REM, and Dream Log phases, including memory statistics and file paths.

```json
{
  "nrem": {
    "episodes": 3,
    "chunks": 45,
    "clusters": 12,
    "created": 8,
    "skipped_dup": 3,
    "docs_created": 1,
    "processed_dates": ["2024-03-15", "2024-03-16", "2024-03-17"],
    "created_ids": ["uuid-1", "uuid-2", "..."]
  },
  "rem": {
    "total_memories": 156,
    "conflicts_found": 4,
    "merged": 2,
    "consolidated": 1,
    "deleted": 3,
    "decayed": 45,
    "soft_deleted": 2,
    "archived": 3
  },
  "dream_log": "/home/user/.dreamer/dream-log/2024-03-18_0300.md",
  "elapsed_seconds": 45.2
}
```

--------------------------------

### OpenClaw Configuration for Dreamer Integration

Source: https://github.com/eesiz/clawdreamer/blob/master/README.md

Configure OpenClaw's plugins to enable the memory-lancedb integration and the session-memory hook. This setup ensures that semantic memories are stored in LanceDB and conversation data is saved to episode files.

```json
{
  "plugins": {
    "slots": {
      "memory": "memory-lancedb"
    },
    "entries": {
      "memory-lancedb": {
        "enabled": true,
        "config": {
          "embedding": {
            "apiKey": "${OPENAI_API_KEY}",
            "model": "text-embedding-3-small"
          },
          "autoCapture": true,
          "autoRecall": true
        }
      }
    }
  },
  "hooks": {
    "internal": {
      "enabled": true,
      "entries": {
        "session-memory": {
          "enabled": true
        }
      }
    }
  }
}
```

--------------------------------

### Initialize Dreamer with Default Path

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Initialize the Dreamer data directory structure and create the LanceDB memories table using the default home path (~/.dreamer).

```bash
python setup.py
```

--------------------------------

### Main Entry Point: dreamer.py

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Orchestrates the complete dream cycle (NREM, REM, Dream Log). Supports running individual phases and provides verbose logging.

```APIDOC
## Main Entry Point: dreamer.py

### Description
Runs the complete dream cycle: NREM (episode processing), REM (memory integration), and Dream Log (report generation). Supports running individual phases and provides verbose logging for debugging.

### Usage
```bash
# Full dream cycle (NREM -> REM -> Dream Log)
python dreamer.py --verbose

# NREM phase only (episode -> semantic conversion)
python dreamer.py --nrem-only

# REM phase only (decay + pruning existing memories)
python dreamer.py --rem-only
```

### Expected JSON Output
```json
{
  "nrem": {
    "episodes": 3,
    "chunks": 45,
    "clusters": 12,
    "created": 8,
    "skipped_dup": 3,
    "docs_created": 1,
    "processed_dates": ["2024-03-15", "2024-03-16", "2024-03-17"],
    "created_ids": ["uuid-1", "uuid-2", "..."]
  },
  "rem": {
    "total_memories": 156,
    "conflicts_found": 4,
    "merged": 2,
    "consolidated": 1,
    "deleted": 3,
    "decayed": 45,
    "soft_deleted": 2,
    "archived": 3
  },
  "dream_log": "/home/user/.dreamer/dream-log/2024-03-18_0300.md",
  "elapsed_seconds": 45.2
}
```
```

--------------------------------

### Initialize Dreamer with Custom Path

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Initialize the Dreamer data directory structure and create the LanceDB memories table using a specified custom home path.

```bash
python setup.py --home /path/to/data
```

--------------------------------

### Run Full Dreamer Cycle

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Execute the complete dream cycle including NREM, REM, and Dream Log phases. Use the --verbose flag for detailed logging.

```bash
python dreamer.py --verbose
```

--------------------------------

### Dreamer Project File Structure

Source: https://github.com/eesiz/clawdreamer/blob/master/DESIGN.md

This outlines the directory and file structure for the Dreamer project. It includes configuration, main entry point, modules for NREM and REM phases, embedding generation, LanceDB operations, LLM interactions, and dream log generation.

```plaintext
dreamer/
├── DESIGN.md          # this document
├── config.py          # configuration
├── dreamer.py         # main entry point
├── nrem.py            # Phase 1: episode -> semantic conversion
├── rem.py             # Phase 2: integration + pruning
├── embedder.py        # OpenAI embedding generation
├── lancedb_store.py   # LanceDB read/write operations
├── llm.py             # LLM calls (summarization/classification)
├── dream_log.py       # dream log generation
└── requirements.txt
```

--------------------------------

### NREM Phase: run_nrem()

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Processes episode markdown files by chunking content, clustering similar chunks, analyzing clusters with LLM, and storing new semantic memories in LanceDB.

```APIDOC
## NREM Phase: run_nrem()

### Description
Processes episode markdown files by chunking content, clustering similar chunks via embedding similarity, analyzing clusters with LLM to extract key facts, and storing new semantic memories in LanceDB after deduplication.

### Usage
```python
from nrem import run_nrem

# Run NREM phase to process episode files
result = run_nrem()
```

### Result Structure
```json
{
    "episodes": 3,           # Number of episode files processed
    "chunks": 45,            # Text chunks extracted
    "clusters": 12,          # Semantic clusters formed
    "created": 8,            # New memories stored
    "skipped_dup": 3,        # Duplicates skipped
    "docs_created": 1,       # Reference docs generated
    "processed_dates": ["2024-03-15", "2024-03-16"],
    "created_ids": ["abc-123", "def-456"]  # IDs for REM phase
}
```
```

--------------------------------

### Configure Telegram Alerts for Dreamer

Source: https://github.com/eesiz/clawdreamer/blob/master/README.md

Sets environment variables to enable and configure Telegram alerts for Dreamer. This bypasses the AI agent and sends error notifications directly to the operator.

```bash
export DREAMER_ALERT_PROVIDER=telegram
export DREAMER_ALERT_TELEGRAM_BOT_TOKEN=123456:ABC-DEF
export DREAMER_ALERT_TELEGRAM_CHAT_ID=your_chat_id
```

--------------------------------

### Dream Log Generation Steps

Source: https://github.com/eesiz/clawdreamer/blob/master/DESIGN.md

This describes the process for generating a summary report of the Dreamer's consolidation activities. The log includes details on new memories created, merged, consolidated, deleted, and any conflicts resolved.

```plaintext
15. Generate summary report -> dream-log/YYYY-MM-DD_HHMM.md
    - New semantic memories created
    - Memories merged/consolidated
    - Memories soft-deleted
    - Conflicts found and resolved
```

--------------------------------

### Add, Update, and Delete Memories

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Manage memories by adding new ones with text, vectors, and metadata, updating their importance or content, or deleting them entirely.

```python
mem_id = add_memory(
    text="User prefers Docker Compose over raw Docker commands.",
    vector=[0.1, 0.2, ...],  # 1536-dim embedding
    importance=0.75,
    category="preference"
)
```

```python
update_importance(mem_id, new_importance=0.85)
```

```python
update_memory_text(
    mem_id,
    new_text="User prefers Docker Compose. Uses nginx as reverse proxy.",
    new_vector=[0.15, 0.22, ...]
)
```

```python
delete_memory(mem_id)
```

--------------------------------

### Load All Memories from LanceDB

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Load all existing memories from the LanceDB vector store using the `load_all_memories` function.

```python
from lancedb_store import (
    load_all_memories,
    add_memory,
    update_importance,
    update_memory_text,
    delete_memory,
    search_similar
)

# Load all memories from database
memories = load_all_memories()
```

--------------------------------

### NREM Phase: Episode to Semantic Memory Conversion Steps

Source: https://github.com/eesiz/clawdreamer/blob/master/DESIGN.md

This outlines the process for converting raw episodic data into structured semantic memories. It involves chunking, embedding generation, clustering, LLM summarization, deduplication against existing memories, and storing new semantic memories with dynamic importance scoring.

```plaintext
1. Load episodes (episodes/*.md)
2. Chunk into semantic units
3. Generate embeddings
4. Cluster similar chunks (cosine similarity > 0.75)
5. LLM summarization per cluster -> extract patterns/principles
6. Dedup check against existing LanceDB memories (similarity > 0.9)
7. Store new semantic memories (dynamic importance scoring)
```

--------------------------------

### Send Error Alerts with send_alert()

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Use this function to send failure notifications to operators via Telegram, Slack, or a generic webhook when errors occur. Configuration is done via environment variables.

```python
from alerts import send_alert

# Automatic alerting on errors (called by dreamer.py)
try:
    # ... dreamer operations
except Exception as e:
    send_alert(e)  # Returns True if delivered
```

```bash
# Configuration via environment variables:
# DREAMER_ALERT_PROVIDER=telegram
# DREAMER_ALERT_TELEGRAM_BOT_TOKEN=123456:ABC-DEF
# DREAMER_ALERT_TELEGRAM_CHAT_ID=your_chat_id

# Or for Slack:
# DREAMER_ALERT_PROVIDER=slack
# DREAMER_ALERT_SLACK_WEBHOOK_URL=https://hooks.slack.com/...

# Or generic webhook (POST JSON):
# DREAMER_ALERT_PROVIDER=webhook
# DREAMER_ALERT_WEBHOOK_URL=https://your-endpoint.com/alert
# Payload: {"source": "dreamer", "text": "...", "error": "..."}
```

--------------------------------

### Run NREM Phase in Python

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Execute the NREM phase using the `run_nrem` function. This processes episode files, chunks content, clusters similar chunks, analyzes clusters with an LLM, and stores new semantic memories in LanceDB after deduplication.

```python
from nrem import run_nrem

# Run NREM phase to process episode files
result = run_nrem()

# Result structure:
# {
#     "episodes": 3,           # Number of episode files processed
#     "chunks": 45,            # Text chunks extracted
#     "clusters": 12,          # Semantic clusters formed
#     "created": 8,            # New memories stored
#     "skipped_dup": 3,        # Duplicates skipped
#     "docs_created": 1,       # Reference docs generated
#     "processed_dates": ["2024-03-15", "2024-03-16"],
#     "created_ids": ["abc-123", "def-456"]  # IDs for REM phase
# }
```

--------------------------------

### REM Phase: run_rem()

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Integrates new memories with existing ones by detecting conflicts, classifying relationships, merging or consolidating, applying importance decay, and archiving processed episodes.

```APIDOC
## REM Phase: run_rem()

### Description
Integrates new memories with existing ones by detecting conflicts, classifying relationships (state_change/different_aspects/unrelated), merging or consolidating as appropriate, applying importance decay to old memories, and archiving processed episodes.

### Usage
```python
from rem import run_rem

nrem_result = {
    "created_ids": ["abc-123", "def-456"],
    "processed_dates": ["2024-03-15"]
}
rem_result = run_rem(nrem_result)
```

### Result Structure
```json
{
    "total_memories": 156,
    "conflicts_found": 4,
    "merged": 2,              # State changes merged
    "consolidated": 1,        # Different aspects combined
    "deleted": 3,             # Memories removed
    "split_created": 1,       # New memories from splits
    "decayed": 45,            # Importance decremented
    "soft_deleted": 2,        # Below threshold, removed
    "archived": 1,            # Episode files archived
    "merge_details": [...],   # Detailed merge info
    "consolidation_details": [...] 
}
```
```

--------------------------------

### Search for Similar Memories

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Find memories similar to a given vector embedding, specifying the number of results and a similarity threshold.

```python
matches = search_similar(
    vector=[0.1, 0.2, ...],
    top_k=5,
    threshold=0.3
)
```

--------------------------------

### Generate Text Embeddings

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Create vector embeddings for text using batch processing for efficiency or single embedding for individual texts. Supports configurable providers like OpenAI, Ollama, and sentence-transformers.

```python
from embedder import embed_texts, embed_single, cosine_similarity

texts = [
    "User prefers dark mode UI",
    "API rate limit is 100 req/min",
    "Database uses PostgreSQL 16"
]
vectors = embed_texts(texts)
```

```python
vector = embed_single("Deploy with Docker Compose")
```

```python
sim = cosine_similarity(vectors[0], vectors[1])
```

--------------------------------

### LanceDB Store Operations

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Handles all vector database operations including loading, adding, updating, searching, and deleting memories with their embeddings.

```APIDOC
## LanceDB Store Operations

### Description
The LanceDB store module handles all vector database operations including loading, adding, updating, searching, and deleting memories with their embeddings.

### Usage
```python
from lancedb_store import (
    load_all_memories,
    add_memory,
    update_importance,
    update_memory_text,
    delete_memory,
    search_similar
)

# Load all memories from database
memories = load_all_memories()
```
```

--------------------------------

### LLM Functions for Memory Analysis

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Utilize LLM-powered functions to analyze clusters of text, classify relationships between memories, merge state changes, and consolidate related information into comprehensive memories.

```python
from llm import (
    analyze_cluster,
    classify_relationship,
    merge_state_change,
    consolidate_aspects,
    llm_call
)

chunks = [
    "Discussed deployment strategy. Decided on Docker Compose.",
    "Will use nginx as reverse proxy for the deployment."
]
existing_files = ["docs/api-guide.md", "skills/deploy/SKILL.md"]

result = analyze_cluster(chunks, existing_files)
```

```python
classification = classify_relationship(
    "Model set to gemini-pro for chat completions",
    "Changed model to claude-3 for better reasoning"
)
```

```python
merged = merge_state_change(
    newer_text="Model is claude-3 for chat completions",
    older_text="Model set to gemini-pro for chat completions"
)
```

```python
consolidated = consolidate_aspects(
    "API endpoint: POST /v1/charges",
    "API rate limit: 100 req/min, auth via Bearer token"
)
```

```python
response = llm_call(
    prompt="Summarize this in one sentence: ...",
    system="You are a helpful assistant.",
    max_tokens=100
)
```

--------------------------------

### Generate Dream Log Report

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Create a detailed markdown report of each dream cycle, documenting memory creation, merging, consolidation, and deletion activities for auditing purposes.

```python
from dream_log import write_dream_log

nrem_result = {
    "episodes": 3, "chunks": 45, "clusters": 12,
    "created": 8, "skipped_dup": 3, "docs_created": 1,
    "processed_dates": ["2024-03-15", "2024-03-16"]
}

rem_result = {
    "total_memories": 156, "conflicts_found": 4,
    "merged": 2, "consolidated": 1, "deleted": 3,
    "decayed": 45, "soft_deleted": 2, "archived": 2,
    "merge_details": [
        {"before": ["old1", "old2"], "after": "merged", "kept_id": "abc"}
    ],
    "consolidation_details": []
}

log_path = write_dream_log(nrem_result, rem_result)
```

--------------------------------

### REM Phase: Semantic Memory Integration and Pruning Steps

Source: https://github.com/eesiz/clawdreamer/blob/master/DESIGN.md

This details the integration and pruning of semantic memories. It includes detecting conflicts between new and existing memories, classifying conflicts, resolving them through merging or consolidation, applying importance decay, soft-deleting low-importance memories, and archiving processed episode files.

```plaintext
8.  Load all semantic memories
9.  Detect conflicts between NEW and EXISTING (O(N*M), not O(N^2))
10. Classify: state_change / different_aspects / unrelated
11. Resolve: merge (state changes) or consolidate (different aspects)
12. Apply importance decay (unrecalled memories fade)
13. Soft-delete memories below threshold (0.15)
14. Archive processed episode files
```

--------------------------------

### Run NREM Phase Only

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Execute only the NREM phase of the dream cycle, which focuses on episode processing and semantic conversion.

```bash
python dreamer.py --nrem-only
```

--------------------------------

### Run REM Phase in Python

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Execute the REM phase using the `run_rem` function, passing the results from the NREM phase. This integrates new memories with existing ones, detects conflicts, classifies relationships, merges or consolidates memories, applies importance decay, and archives processed episodes.

```python
from rem import run_rem

# Run REM phase with NREM results
nrem_result = {
    "created_ids": ["abc-123", "def-456"],
    "processed_dates": ["2024-03-15"]
}
rem_result = run_rem(nrem_result)

# Result structure:
# {
#     "total_memories": 156,
#     "conflicts_found": 4,
#     "merged": 2,              # State changes merged
#     "consolidated": 1,        # Different aspects combined
#     "deleted": 3,             # Memories removed
#     "split_created": 1,       # New memories from splits
#     "decayed": 45,            # Importance decremented
#     "soft_deleted": 2,        # Below threshold, removed
#     "archived": 1,            # Episode files archived
#     "merge_details": [...],   # Detailed merge info
#     "consolidation_details": [...] 
# }
```

--------------------------------

### Run REM Phase Only

Source: https://context7.com/eesiz/clawdreamer/llms.txt

Execute only the REM phase of the dream cycle, which handles memory decay and pruning of existing memories.

```bash
python dreamer.py --rem-only
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.