openclaw-workspace-2026/architecture/NOCODB_SETUP.md

NocoDB Setup for Self-Correcting Memory

Current Configuration

Agent Base ID: pedwxnsn51vxdq2
Memory Table ID: mwdr70ocb7iwnrg

Issue

The automation token (svc-automation role) cannot access the Agent base. This is a permissions issue.

Solution Options

Option 1: Grant Automation Role Access to Agent Base (Recommended)

In NocoDB web UI:

1. Go to Agent base settings
2. Add role svc-automation with Read/Write permissions
3. Or add user associated with automation role

Option 2: Create Tables with Admin Token

Use a token with broader permissions to create these tables in the Agent base:

Table: corrections

CREATE TABLE corrections (
    id VARCHAR(50) PRIMARY KEY,
    pattern VARCHAR(500) NOT NULL,
    exclude_pattern VARCHAR(500),
    severity VARCHAR(20) NOT NULL,
    message VARCHAR(500) NOT NULL,
    suggestion VARCHAR(500),
    blocking BOOLEAN DEFAULT TRUE,
    auto_detected BOOLEAN DEFAULT FALSE,
    manual_entry BOOLEAN DEFAULT FALSE,
    hit_count INTEGER DEFAULT 0,
    last_triggered TIMESTAMP,
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
    updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
    context VARCHAR(200)
);

Table: memory_preferences

CREATE TABLE memory_preferences (
    id VARCHAR(50) PRIMARY KEY,
    category VARCHAR(50) NOT NULL,
    key VARCHAR(100) NOT NULL,
    value TEXT NOT NULL,
    importance INTEGER DEFAULT 5,
    confidence FLOAT DEFAULT 1.0,
    tags TEXT, -- JSON array
    access_count INTEGER DEFAULT 0,
    last_accessed TIMESTAMP,
    confirmed_count INTEGER DEFAULT 0,
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
    updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);

Table: memory_episodes

CREATE TABLE memory_episodes (
    id VARCHAR(50) PRIMARY KEY,
    date DATE NOT NULL,
    summary TEXT NOT NULL,
    details TEXT,
    project VARCHAR(100),
    outcomes TEXT, -- JSON
    corrections_triggered TEXT, -- JSON array
    vector_embedding TEXT, -- Future use
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);

Table: memory_decisions

CREATE TABLE memory_decisions (
    id VARCHAR(50) PRIMARY KEY,
    date DATE NOT NULL,
    project VARCHAR(100),
    decision TEXT NOT NULL,
    alternatives TEXT, -- JSON
    rationale TEXT NOT NULL,
    status VARCHAR(20) DEFAULT 'active',
    reversed_by VARCHAR(50),
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);

Table: validation_runs

CREATE TABLE validation_runs (
    id VARCHAR(50) PRIMARY KEY,
    session_id VARCHAR(50),
    timestamp TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
    input_length INTEGER,
    output_length INTEGER,
    violations_found INTEGER DEFAULT 0,
    new_patterns_detected INTEGER DEFAULT 0,
    corrections_auto_stored INTEGER DEFAULT 0,
    processing_time_ms INTEGER,
    blocked BOOLEAN DEFAULT FALSE,
    workflow VARCHAR(50)
);

Option 3: Use Existing Memory Table

Update memory-service.js to use the existing mwdr70ocb7iwnrg table:

1. Add these columns to the existing table (via NocoDB UI):

   • pattern (LongText)
   • severity (SingleSelect: error/warning/auto-correct)
   • message (LongText)
   • blocking (Checkbox)
   • auto_detected (Checkbox)
   • context (LongText)
   • created_at (DateTime)

2. Update the service to use these columns

Next Steps

1. Choose one of the options above
2. Update architecture/memory-service.js with actual table IDs
3. Test connection: node architecture/test-memory-connection.js

Current Fallback

Until NocoDB is connected, the system uses JSON files:

• memory/corrections/_index.json - Correction patterns
• memory/items/*.json - Preferences
• memory/*.md - Episodes

This works but lacks:

• Concurrent access
• Query capabilities
• Validation logging
• Statistics tracking