feat(NGUI-474/33/42): Update New Modular Architecture for NGUI E2E Server

[Hrithik-Gavankar]

Migrated NGUI E2E server to Lightrail/LlamaStack with modular architecture.

Issues Resolved
✅ NGUI-442 - Fixed empty data_path and data arrays for nested/complex structures
✅ NGUI-433 - Added ability to mock backend data via REST API

Key Changes:

1. Lightrail Compatibility

• Migrated from LangGraph to LlamaStack
• Updated to next-gen-ui-agent & next-gen-ui-llama-stack v0.3.0+ (fixes empty data bug)
• Added Lightrail deployment configuration

2. Modular Architecture

• Inspired by PR #220:

app/
├── agents/ # NGUI agent setup
├── data_sources/ # Inline, default, filtering
├── routes/ # API endpoints
├── utils/ # Validation, logging, responses
└── config.py # Configuration

3. New Features

• Default data fallback - Uses movies_data.json when no data provided
• Smart filtering agent - Intelligently filters data based on prompts
• Enhanced error handling - Standardized error codes with suggestions

Testing:

Local deployment

lightrail-local-cli build && lightrail-local-cli start

Test with default data

curl -X POST http://localhost:8080/generate \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Show me some movies"}'

# Test with custom data (mock)
curl -X POST http://localhost:8080/generate \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Show products", "data": [{"name":"Laptop","price":999}]}'

[velias]

Thanks, I like this modularization of the app, hope it somehow corresponds to the @jschuler proposal from his debug view PR.
I added few proposals and questions.

[jschuler]

Hi @Hrithik-Gavankar, this branch is a bit out of date and needs a merge from main branch. Also, I didn't debug too much but I think it breaks the current way of running the test server locally, which is:

• have a tests/ngui-e2e/server/.env file with contents from models.corp, example:

# gemini
# https://gitlab.cee.redhat.com/models-corp/user-documentation/-/blob/main/models/gemini.md
LLM_MODEL=gemini-2.0-flash
LLM_BASE_URL=https://gemini--apicast-production.apps.int.stc.ai.prod.us-east-1.aws.paas.redhat.com:443/v1beta/openai
LLM_API_KEY=<KEY HERE>

• start tests/ngui-e2e/client

npm run dev

• start tests/ngui-e2e/server

./start_server.sh

can you try and run this as baseline, to ensure it works and also after you add lightrail support?

I would also suggest to make minimal lightrail changes first by:

1. Create tests/ngui-e2e/server/.light folder and files
2. Update config tests/ngui-e2e/server/config.py to load from either set of variables
3. Update BUILD file tests/ngui-e2e/server/BUILD

Once we have this minimally added we can create followup PR with other improvements? WDYT

[velias]

Why are the same files and folder structures directly in the /server folder, but also in the /server/app subfolder? This seems suspicious to me 😉

[Hrithik-Gavankar]

Why are the same files and folder structures directly in the /server folder, but also in the /server/app subfolder? This seems suspicious to me 😉

Thanks @velias for noticing this 👍, Those root-level modules were leftover from the migration and aren't used - the application only imports from app/ to align with Lightrail's structure. Removed the dead code in the latest commit.

[velias]

Thanks, looks good now

[jschuler]

I would have liked to keep this. makes it easy to start the server locally

[jschuler]

this was a useful endpoint for me