AI Agent that updates its own rules to modify behavior

Video walkthrough

https://www.youtube.com/watch?v=OwIFK-r-NtQ

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.

You can tell it things like "Next time, use present tense when talking about this subject" and it will use a tool to save this as a rule, then receive that instruction in all future iterations.

How to start using it

Option 1: With a Postgres database (e.g., Supabase) 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 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 Google Sheet Setup: Create a Google Sheet with columns for rule_text and agent. 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