There's a Grafana dashboard running. You can log into http://localhost:3000 and look at the TrustGraph dashboard. The login is user admin, password admin, and you can skip the password change because the Grafana dashboard isn't available to anyone unless they use an SSH tunnel like you did earlier.
The deploy leaves a single dashboard which should be listed on the home page, if not, navigate into Dashboards and find the Overview dashboard.
Go to http://localhost:8888 to see the workbench.
This sets up a Python 3.12 env directory, which we can put TrustGraph things into...
mkdir -p ~/trustgraph/env
python3.12 -m venv ~/trustgraph/env
Configure to use this new environment. This command need to be run in any terminal window you want to run TrustGraph commands.
. ~/trustgraph/env/bin/activate
And then you can install some TrustGraph packages. The CLI tools may be useful...
pip install trustgraph-cli==0.20.11
There's a command-line arg which talks straight to the LLM, so we can
check that LLM services are working. tg-invoke-llm uses TrustGraph
services to respond to a prompt. Two parts are provided: a system prompt
and a user prompt. Try:
% tg-invoke-llm 'Be helpful' 'Describe San Francisco in 5 words'
Hilly, tech-driven, diverse, scenic, vibrant.
Cognitive cores are a quick way to load pre-generated data into TrustGraph, ready for prompting.
Download the an NTSB accident dataset...
curl -o- https://raw.githubusercontent.com/trustgraph-ai/catalog/refs/heads/master/v3/ntsb.core.gz > ntsb.core.gz
gunzip ntsb.core.gz
And then you can load the core into TrustGraph...
tg-load-kg-core -i ntsb.core
You get a bit of debug output showing counts of loaded objects. The core takes a good few minutes to load, but after about a minute you're good to start interacting with the data that's loaded on the workbench.
But you can also try a GraphRAG query on the command-line...
tg-invoke-graph-rag -q 'Describe an incident where an AA-5 was involved in a mid-air collision'
and
tg-invoke-graph-rag -q 'Who was the investigator for NTSB incident 199414'
You loaded the data into processing queues. There's a pub/sub backlog chart which shows how much data there is waiting to be processed so you can monitor load progress.
Things you can do...
- Ask questions in the Chat window, these are answered using the Graph RAG engine.
- Search for terms in the search window, this finds nodes in the knowledge graph.
- Click on resultant nodes in the search window, which shows you graph triples
- Click GRAPH VIEW to view a 3D knowledge graph of the triples you have discovered.
- This doesn't show the whole graph, you can click on nodes to add more information to the graph.
- Use the pointer to rotate space and move around the 3D space.
Next: Working with Memgraph