Rebuilding a No-Code LLM Workflow in n8n with Supabase Retrieval
Rebuilt an Australian client's no-code LLM workflow in n8n, with retrieval over a Supabase knowledge base, and tested the answers side by side against the original.
- Client
- Australia-based client
- Industry
- AI knowledge assistant
- Timeline
- Jun to Sep 2025
- Engagement
- Fixed price
The system, step by step
In this case study, we cover moving an AI workflow from a no-code platform to n8n and Supabase. The aim was the same behaviour, on infrastructure the client controls.
Question
Asked of the assistant
Supabase retrieval
Filtered by project stage
LLM answer
Built in n8n
Side-by-side test
Compared to the original
What the brief demanded
Matching a working system closely needed careful testing:
The client
The client was based in Australia and ran an AI assistant on a hosted no-code platform. They wanted the same answers from a workflow they could own and extend.
- Reproducing the original retrieval behaviour over the client's knowledge base.
- Filtering results by project stage so each answer draws on the right material.
- Locating source files in Supabase storage from knowledge base records.
- Comparing outputs against the original platform on the same questions.
Built and handed over
We rebuilt the workflow in n8n with Supabase as the knowledge base, and worked through a shared test workbook with the client to compare the responses.
- An n8n workflow with retrieval over a Supabase knowledge base.
- Project-stage filtering in the retrieval step.
- A side-by-side test process against the original no-code workflow.
The project ran from June to early September 2025 as a fixed-price engagement and was paid in full on completion, with the client reviewing and refining the workflow through July.
- n8n
- Supabase
- Retrieval-augmented generation
- A large language model
Want a system like this?
Tell us the process that costs your team the most time. On a 30-minute call we will say what we would build and what it would take.
Book a Call