zero-cost-ops

Workflows

Two n8n workflows are included. Together they handle auto-ingestion, AI classification, and timeline organization of your system’s metadata.

---

Overview

Workflow	Trigger	What it does
WF-01 — Ingestion Pipeline	Webhook (POST)	Receives content → filters noise → classifies with Claude → stores in Supabase → triggers cross-linking
WF-04 — Timeline Engine	Webhook or manual	Scans all entries → clusters by time period → detects narrative arc → bulk-updates timeline_group field

---

WF-01: Ingestion Pipeline

File: WF-01-ingestion-pipeline.json

What it does

Receives any content via webhook and processes it into a structured metadata entry.

Webhook → Extract Content → Filter Messages → Claude Classify → Parse → Supabase Insert → Trigger WF-02 → Respond

Node breakdown

1. Webhook Trigger — Listens on POST /webhook/wf01-ingest. Accepts JSON with text, source, user_id, timestamp, and optional metadata.

2. Extract Content — Normalizes the payload. Extracts message_text, source, user_id, and timestamp regardless of how the sender structured the POST body.

3. Filter Messages — Drops messages shorter than 10 characters and slash commands. Prevents noise from getting classified.

4. Claude API Classification — Sends the message to claude-sonnet-4-20250514. The prompt extracts:
   • tags (3–5 keywords)
   • entities (people, companies, concepts)
   • summary (1–2 sentences)

   • narrative_role (symbol

     system

     pilot

     origin

     exploration

     setup

     conflict

     realisation

     integration)

   • date_estimated (YYYY-MM-DD or YYYY)

   • timeline_group (1974-1990

     2019-2023

     2024-2026)

5. Parse Classification — Extracts JSON from Claude’s response. Handles markdown code blocks gracefully. Generates a unique file_name (auto_{source}_{timestamp}).

6. Supabase Insert — Inserts the full classified entry into the metadata table.

7. Trigger WF-02 — POSTs to the cross-linking workflow endpoint (if you have one). Remove or modify this node if you don’t use WF-02.

8. Respond Success — Returns { "status": "success", "file_name": "..." } to the original caller.

Usage

Send a POST request to your n8n instance:

curl -X POST https://YOUR-N8N-INSTANCE.onrender.com/webhook/wf01-ingest \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Deployed new RLS policies to Supabase. All security checks passing.",
    "source": "github",
    "user_id": "your-user-id"
  }'

Customization

• Change the prompt: Edit the messages body parameter in the Claude API Classification node to extract different fields.
• Add more fields: Add new columns to the Supabase Insert node’s mapping after adding them to the metadata table.
• Different LLM: Replace the HTTP Request node with any other LLM API — the prompt structure works with GPT-4, Gemini, etc.
• Different sources: The source field is free-text. Use it to track where content came from (slack, github, manual, twitter, etc.).

---

WF-04: Timeline Engine

File: WF-04-timeline-engine.json

What it does

Scans all entries in your metadata table and assigns them to temporal clusters based on date_estimated.

Webhook → Fetch All Entries → Temporal Clustering → Narrative Arc Detection → Split in Batches → Extract Entry → Update Timeline Group → Respond

Node breakdown

1. Manual Trigger — Webhook on POST /webhook/wf04-timeline. Trigger manually or on a cron schedule (weekly recommended).

2. Fetch All Entries — Retrieves all rows from the metadata table via Supabase.

3. Temporal Clustering — For each entry, parses date_estimated and assigns a timeline_group:
   • 1974-1990 — Origin/early period
   • 2019-2023 — Exploration period
   • 2024-2026 — Current/integration period

4. Narrative Arc Detection — Groups entries by timeline period and maps the dominant narrative role for each. Returns an analysis summary.

5. Split in Batches — Processes entries one at a time to avoid Supabase rate limits.

6. Extract Entry — Pulls the current entry from the batch for processing.

7. Update Timeline Group — Updates the timeline_group field in Supabase for the current entry.

8. Respond Summary — Returns a summary with total entries updated and the narrative arc analysis.

Usage

curl -X POST https://YOUR-N8N-INSTANCE.onrender.com/webhook/wf04-timeline

Customization

• Change time periods: Edit the date ranges in the Temporal Clustering function node.
• Add new periods: Add keys to the groups object and the corresponding year range checks.
• Scheduled runs: Add a Schedule Trigger node to run weekly automatically.
• Different classification: The narrative_role values are custom — change them to match your project’s taxonomy.

---

How to Import

1. In your n8n instance, go to Workflows > Import from File
2. Select WF-01-ingestion-pipeline.json
3. Update credentials:
   • Supabase: Go to the Supabase Insert node, click “Create New Credential”, enter your project URL and service role key
   • Anthropic: The Claude API node uses predefinedCredentialType: anthropicApi — add your API key in n8n Settings > Credentials
4. Update the Trigger WF-02 node URL to point to your n8n instance (or delete it if unused)
5. Activate the workflow
6. Repeat for WF-04

---

Environment Variables for n8n on Render

# Required
SUPABASE_URL=https://YOUR-PROJECT.supabase.co
SUPABASE_SERVICE_ROLE_KEY=your-service-role-key

# Optional — set in n8n credentials manager instead
ANTHROPIC_API_KEY=your-anthropic-key

# n8n configuration
N8N_BASIC_AUTH_ACTIVE=true
N8N_BASIC_AUTH_USER=admin
N8N_BASIC_AUTH_PASSWORD=your-secure-password
N8N_HOST=0.0.0.0
N8N_PORT=5678
WEBHOOK_URL=https://YOUR-N8N-INSTANCE.onrender.com

---

Troubleshooting

Workflow not triggering

• Check the webhook URL — must include https:// and your exact Render domain
• Verify the workflow is active (toggle in top-right of workflow editor)
• Check Executions tab for error details

Claude API returning non-JSON

• The Parse Classification node handles markdown code blocks. If Claude returns something unexpected, add a console.log(content) to the function to see raw output.
• Claude occasionally adds explanatory text — the JSON extraction handles \``json` wrappers.

Supabase insert failing

• Verify your service role key (not anon key) is in the Supabase credentials
• Check that RLS policies allow authenticated inserts on the metadata table
• Ensure all required columns exist — run schema/supabase-schema.sql first

n8n sleeping on Render free tier

• Free tier instances sleep after 15 minutes of inactivity
• Add a UptimeRobot monitor to ping https://YOUR-N8N.onrender.com/healthz every 5 minutes
• This keeps the instance warm at no cost

This site is open source. Improve this page.