Skip to content
NEWMarketplace and the trig8n Agent: sell what you build, and get plans and changes that are tested before you see them. See what is new

Library / AI Agents & Assistants

AI Agents & Assistants

Build a Claude Assistant That Rewrites Its Own Rules

When a chat message arrives, a Claude assistant stores and updates its own behavior rules in a Postgres database and uses Google Sheets as a tool.

What it does

## Summary of agent This agent can write and rewrite its own rules, allowing you to mold its behavior. It receives rules from a database as system instructions and has tools to create, edit, or delete them. This is a great baseline for new agent builds. ## How to start using it ### Option 1: With a Postgres database (e.g., Supabase) 1. Supabase Schema: Create a table (e.g., agent_rules) with the following columns: * id: bigint (Primary Key, auto-incrementing) * created_at: timestamp with time zone (Default: now()) * rule_text: text * agent: text 2. Workflow Updates: * Update the Postgres credentials in the "Get rules from database," "Insert rule into database," and "Execute query on rule database" nodes. * Update the agent value (currently 'TestAgent') in the "Get rules from database" and "Insert rule into database" nodes if you want a different agent name. * Update the Anthropic API credentials. ### Option 2: With Google Sheets 1. Google Sheet Setup: Create a Google Sheet with columns for rule_text and agent. 2. Workflow Updates: * Example Google Sheets nodes are included. You will need to: * Connect your Google Sheets credentials. * Select your Google Sheet (with rule_text and agent columns) in all relevant Google Sheets nodes. * Replace the existing Postgres nodes ("Get rules from database", "Insert rule into database", "Execute query on rule database") with the configured Google Sheets nodes. * Update the agent value (currently 'TestAgent') in the Google Sheets nodes if you want a different agent name. * Update the Anthropic API credentials. * Agent Instructions: Update the agent's system message and remove the database schema section as it is no longer relevant

The steps

  1. Ultimate Assistant1
  2. Chat History1
  3. Anthropic Chat Model
  4. Get rules from database
  5. Merge
  6. Aggregate
  7. Insert rule into database
  8. Execute query on rule database
  9. When chat message received
  10. Get rules from sheets
  11. Add rule to sheet
  12. Update rule in sheet
  13. Delete rule in sheet

Similar automations