Setting Up n8nClaw: Self-Hosted AI Assistant in n8n

February 24th, 2026 by Brian

Setting Up n8nClaw: Self-Hosted AI Assistant in n8n

n8nClaw is a self-hosted AI assistant that runs entirely inside n8n workflows. Unlike API-based solutions, this gives you full control over your AI agent’s behavior, memory, and data.

What You’re Building

A Claude-powered assistant that:

Responds across multiple channels (Telegram, WhatsApp, Gmail)
Maintains persistent memory using vector embeddings
Manages tasks with specialized sub-agents
Runs autonomous operations on schedule
Learns about you over time

Think of it as your own personal Claw Marketing agent (from our previous workflow) but with multi-channel support and long-term memory.

Prerequisites

Required Services

You’ll need accounts and API keys for:

Essential:

n8n (self-hosted or cloud)
OpenRouter (Claude API access)
Supabase (vector database for memory)
PostgreSQL (chat history storage)
OpenAI (embeddings API)
Telegram Bot Token

Optional:

Evolution API (WhatsApp integration)
Tavily (web research)
Google Workspace (Gmail, Docs, Gemini)

Cost Estimate

n8n: Free (self-hosted) or $20/month (cloud)
OpenRouter: ~$0.003/1K tokens (Claude Sonnet 4.5)
Supabase: Free tier works for most use cases
OpenAI embeddings: ~$0.0001/1K tokens
Total: Expect $10-30/month depending on usage

Technical Requirements

Basic n8n workflow knowledge
Comfort with JSON configuration
Access to manage credentials and webhooks
Ability to run SQL queries (for Supabase setup)

Installation Steps

Step 1: Get the Workflow

1. Go to github.com/shabbirun/n8nclaw

2. Download `n8nClaw.json` from the repo

3. In n8n, go to Workflows → Import from File

4. Select the downloaded JSON file

You’ll see a complex workflow with multiple nodes. Don’t panic – we’ll configure it step by step.

Step 2: Set Up Data Tables

n8nClaw uses three n8n data tables to track state.

Create Table 1: User Profiles

Name: `n8nclaw_users`
Schema:
`user_id` (string) – Telegram/WhatsApp user ID
`name` (string) – User’s name
`profile` (string) – Dynamic personality/context
`created_at` (datetime)
`updated_at` (datetime)

Create Table 2: Tasks

Name: `n8nclaw_tasks`
Schema:
`task_id` (string, auto-generated)
`user_id` (string)
`description` (string)
`status` (string) – pending/in_progress/completed
`created_at` (datetime)

Create Table 3: Subtasks

Name: `n8nclaw_subtasks`
Schema:
`subtask_id` (string, auto-generated)
`task_id` (string)
`description` (string)
`assigned_to` (string) – Which sub-agent handles it
`status` (string)
`created_at` (datetime)

Copy the table IDs – you’ll need them for workflow configuration.

Step 3: Configure Supabase Vector Store

n8nClaw uses Supabase with pgvector for semantic memory.

In Supabase SQL Editor:

-- Enable pgvector extension
create extension if not exists vector;

-- Create memory table
create table n8nclaw_memory (
  id bigserial primary key,
  user_id text not null,
  content text not null,
  embedding vector(1536),
  created_at timestamp with time zone default now()
);

-- Create index for similarity search
create index on n8nclaw_memory using ivfflat (embedding vector_cosine_ops)
  with (lists = 100);

Get your Supabase connection details:

Project URL: `https://[project-id].supabase.co`
Service role key: Found in Settings → API
Database password: From your project setup

Step 4: Set Up Credentials in n8n

Go to Settings → Credentials and create:

OpenRouter API

Credential type: HTTP Request
Authentication: Generic Credential Type
Add header: `Authorization: Bearer YOUR_OPENROUTER_KEY`

Supabase

Host: Your project URL
Database: `postgres`
User: `postgres`
Password: Your database password
Port: `5432`
SSL: Enabled

OpenAI API

API Key: Your OpenAI key (for embeddings only)

Telegram Bot

Access Token: Get from @BotFather on Telegram
Commands:
`/start` – Initialize conversation
`/help` – Show available commands
`/memory` – Show recent memory

PostgreSQL (for chat history)

Host: Your Postgres instance
Database name
User/password
SSL enabled

Step 5: Configure Workflow Variables

Open the imported workflow and find these nodes to update:

Main Agent Node:

Replace `YOUR_USERNAME` with your actual username
Replace `YOUR_TELEGRAM_CHAT_ID` with your Telegram chat ID (get from @userinfobot)
Update table IDs to match your n8n tables

Memory Workflow:

Update Supabase credentials
Set embedding model: `text-embedding-3-small`
Configure retrieval limit: 5 relevant memories

Sub-Agent Nodes:

Research Agent: Add Tavily API key if using
Email Manager: Configure Gmail OAuth
Document Manager: Set up Google Docs access

Heartbeat Schedule:

Default: Every hour
Adjust based on your needs (don’t go below 15 min to avoid rate limits)

Step 6: Test Each Component

Before activating, test each workflow segment:

Test 1: Telegram Connection

1. Send a message to your bot

2. Check if webhook receives it

3. Verify chat history saves to Postgres

Test 2: Claude Response

1. Send `/help` command

2. Should get response from Claude

3. Check OpenRouter dashboard for API call

Test 3: Memory Retrieval

1. Have a conversation about a topic

2. Wait for memory workflow to run (check schedule)

3. Ask about the same topic later – should recall context

Test 4: Task Management

1. Ask agent to “create a task to research AI tools”

2. Check `n8nclaw_tasks` table for new entry

3. Should assign to appropriate sub-agent

Step 7: Activate and Monitor

1. Click “Activate” in the workflow

2. Enable all sub-workflows (memory, heartbeat, sub-agents)

3. Monitor execution logs for first 24 hours

4. Check error logs in n8n executions panel

Configuration Tips

Optimize Memory Usage

The default 15-message context window works well but you can adjust:

Increase to 25 for longer conversations
Decrease to 10 for faster responses
Memory workflow runs hourly – change to daily if you want less frequent summarization

Customize Agent Personality

Edit the system prompt in the Main Agent node:

You are a helpful AI assistant with access to:
- Long-term memory about the user
- Task management capabilities
- Specialized sub-agents for research, email, and documents

Your personality: [professional/casual/technical - customize here]

When you don't know something, say so. Don't make up information.

Set Up Sub-Agent Routing

The workflow routes tasks to sub-agents by complexity:

Haiku – Quick tasks, simple questions
Sonnet – Research, analysis, writing
Opus – Complex reasoning, long documents

Adjust routing logic in the Task Manager node based on keywords or task type.

Common Issues

Memory Not Retrieving

Problem: Agent doesn’t recall past conversations

Fix:

1. Check if memory workflow is running (look at execution history)

2. Verify embeddings are being created (check OpenAI usage)

3. Confirm Supabase vector table has entries: `select count(*) from n8nclaw_memory;`

4. Test similarity search manually in Supabase

Telegram Not Responding

Problem: Bot doesn’t reply to messages

Fix:

1. Confirm webhook is set up correctly in Telegram

2. Check n8n webhook URL is publicly accessible

3. Verify bot token is valid (test with Telegram API)

4. Look for errors in n8n execution logs

High API Costs

Problem: OpenRouter bills are higher than expected

Fix:

1. Reduce context window from 15 to 10 messages

2. Switch some sub-agents from Sonnet to Haiku

3. Disable heartbeat or reduce frequency

4. Add rate limiting to webhook triggers

Tasks Not Completing

Problem: Tasks stuck in “in_progress” status

Fix:

1. Check sub-agent workflows are active

2. Verify credentials for external services (Gmail, Tavily, etc.)

3. Add timeout handling to sub-agent nodes (max 5 minutes)

4. Review execution logs for specific error messages

Next Steps

Once n8nClaw is running:

1. Extend Channels – Add Discord, Slack, or custom webhooks

2. Build Custom Sub-Agents – Create specialized workers for your use case

3. Enhance Memory – Implement semantic search with filters (by date, topic, etc.)

4. Add Tools – Integrate with your CRM, calendar, or project management

5. Scale Up – Move from Supabase free tier to paid for production use

Maintenance

Weekly:

Review execution logs for errors
Check API usage and costs
Monitor Postgres database size

Monthly:

Clean up old chat history (keep last 90 days)
Review and optimize memory embeddings
Update Claude model if newer version available

As Needed:

Rebuild memory index if search quality degrades
Adjust sub-agent routing based on actual task patterns
Fine-tune personality prompts based on user feedback

Resources

Related Workflows

→ DataForSEO + n8n Integration – Another n8n automation workflow
→ Claw Marketing Setup – Autonomous AI agent concepts

—

This workflow requires intermediate n8n knowledge. Budget 2-3 hours for initial setup. Once running, maintenance is minimal (< 30 min/week).

Posted in Workflows

No Comments

Setting Up n8nClaw: Self-Hosted AI Assistant in n8n

Setting Up n8nClaw: Self-Hosted AI Assistant in n8n

What You’re Building

Prerequisites

Required Services

Cost Estimate

Technical Requirements

Installation Steps

Step 1: Get the Workflow

Step 2: Set Up Data Tables

Step 3: Configure Supabase Vector Store

Step 4: Set Up Credentials in n8n

Step 5: Configure Workflow Variables

Step 6: Test Each Component

Step 7: Activate and Monitor

Configuration Tips

Optimize Memory Usage

Customize Agent Personality

Set Up Sub-Agent Routing

Common Issues

Memory Not Retrieving

Telegram Not Responding

High API Costs

Tasks Not Completing

Next Steps

Maintenance

Resources

Related Workflows

Leave a Reply Cancel reply

Resources