Can't type in web UI Vector Search or Assistant - Error: TypeError: Cannot read properties of undefined (reading 'map')

[hdotking-materra]

UI search fails (TypeError: Cannot read properties of undefined (reading 'map')) and tg-show-graph returns no output

Summary

• Submitting text in Vector Search or Assistant in Workbench UI errors with:

  TypeError: Cannot read properties of undefined (reading 'map')

• CLI graph inspection returns no output:

  tg-show-graph
  # (nothing prints)

• Processors are healthy, flows look correct, and sample documents download + submit successfully.

Environment

• macOS Sequoia 15.5 (24F74) (Apple Silicon)
• Docker Desktop (Linux VM kernel in logs: 6.10.14-linuxkit)
• Compose images (from docker-compose.yaml):
  • trustgraph-flow:1.2.20
  • trustgraph-mcp:1.2.20
  • workbench-ui:1.0.6
  • apachepulsar/pulsar:4.0.5
  • cassandra:4.1.9
  • qdrant:v1.14.1
  • minio:RELEASE.2025-06-13T11-33-47Z
  • prom/prometheus:v2.53.5
  • grafana:12.0.2
• CLI
  • uv installed
  • trustgraph-cli==1.3.9 (installed via uv tool install and also tested with uvx --from trustgraph-cli …)
  • trustgraph-base==1.3.9
• TrustGraph version: 0.18.14

What works

Processors are healthy:

tg-show-processor-state

chunker                        💚
metering                       💚
text-completion-rag            💚
text-completion                💚
kg-store                       💚
kg-extract-relationships       💚
store-triples                  💚
kg-extract-definitions         💚
kg-manager                     💚
query-triples                  💚
config-svc                     💚
pdf-decoder                    💚
metering-rag                   💚
graph-rag                      💚
embeddings                     💚
store-graph-embeddings         💚

Flows present:

tg-show-flows

+-------+------------------------------------------------------------+
| id    | default                                                    |
| class | document-rag+graph-rag                                     |
| desc  | Default processing flow                                    |
| queue | document-load: persistent://tg/flow/document-load:default  |
|       | text-load: persistent://tg/flow/text-document-load:default |
+-------+------------------------------------------------------------+

Sample documents load & submit successfully:

tg-load-sample-documents (excerpt)

Report of the Presidential Commission on the Space Shuttle Challenger Accident, Volume 1 :
  downloading... done. adding... successful.
A Concise Dictionary of Old Icelandic :
  downloading... done. adding... successful.
Annual threat assessment of the U.S. intelligence community - March 2025 :
  downloading... done. adding... successful.
The Role of Intelligence and State Policies in International Security :
  downloading... done. adding... successful.
Beyond the vigilant state: globalisation and intelligence :
  downloading... done. adding... successful.

What doesn’t work

• CLI: tg-show-graph prints no output.

• UI: Typing in Vector Search or Assistant throws a frontend runtime error:

  Error: TypeError: Cannot read properties of undefined (reading 'map')
  

Screenshot

Steps to reproduce

1. Start the stack with the compose file pinning the images listed above (Pulsar, Cassandra, Qdrant, MinIO, Prometheus, Grafana, Workbench UI, TrustGraph services).
2. Install CLI (either of the below reproduces):
   • uv tool install trustgraph-cli then tg-show-processor-state
   • or uvx --from trustgraph-cli tg-show-processor-state
3. Load sample docs: tg-load-sample-documents.
4. Open Workbench UI at http://localhost:8888.
5. Go to Vector Search or Assistant, type any query → observe error.
6. In a terminal, run tg-show-graph → observe no output.

Expected behaviour

• tg-show-graph prints triples / graph contents.
• UI search panes return results (or at minimum, no crash if results are empty), i.e., graceful handling with an empty list.

Actual behaviour

• tg-show-graph prints nothing.
• UI shows TypeError: Cannot read properties of undefined (reading 'map') on query submission.

Additional diagnostics

• Gateway /metrics responds (Prometheus scraping OK).
• Broker and services report healthy in CLI.
• Available to provide on request:
  • API Gateway logs around the failing UI requests
  • Browser DevTools → Network response payload for the failing search call
  • Frontend console stack trace
  • tg-show-library-processing output
  • Cassandra counts (e.g., triples_by_subject)

Apple Silicon (arm64) note

On macOS (M-series), I hit architecture mismatches when pulling/running some images
(e.g. errors like no matching manifest for linux/arm64/v8 or exec format error).

Workaround I used: force linux/amd64 for the images that don’t publish arm64 builds by adding
a docker-compose.override.yaml:

# docker-compose.override.yaml
# Define an anchor for the amd64 platform once, then merge it into each TG service.
x-amd64: &amd64 { platform: linux/amd64 }

services:
  agent-manager:
    <<: *amd64
  api-gateway:
    <<: *amd64
  chunker:
    <<: *amd64
  config-svc:
    <<: *amd64
  document-embeddings:
    <<: *amd64
  document-rag:
    <<: *amd64
  embeddings:
    <<: *amd64
  graph-embeddings:
    <<: *amd64
  graph-rag:
    <<: *amd64
  init-trustgraph:
    <<: *amd64
  kg-extract-agent:
    <<: *amd64
  kg-extract-definitions:
    <<: *amd64
  kg-extract-objects:
    <<: *amd64
  kg-extract-relationships:
    <<: *amd64
  kg-manager:
    <<: *amd64
  kg-store:
    <<: *amd64
  librarian:
    <<: *amd64
  mcp-tool:
    <<: *amd64
  metering:
    <<: *amd64
  metering-rag:
    <<: *amd64
  pdf-decoder:
    <<: *amd64
  prompt:
    <<: *amd64
  prompt-rag:
    <<: *amd64
  query-doc-embeddings:
    <<: *amd64
  query-graph-embeddings:
    <<: *amd64
  query-triples:
    <<: *amd64
  store-doc-embeddings:
    <<: *amd64
  store-graph-embeddings:
    <<: *amd64
  store-objects:
    <<: *amd64
  store-triples:
    <<: *amd64
  text-completion:
    <<: *amd64
  text-completion-rag:
    <<: *amd64
  workbench-ui:
    <<: *amd64
  mcp-server:
    <<: *amd64

Notes:

• In Docker Desktop, I also enabled Settings → Features in development → “Use Rosetta for x86/amd64 emulation on Apple Silicon” (when available).
• You can verify image architecture with:
  • docker buildx imagetools inspect <image:tag>
  • docker image inspect --format '{{.Architecture}}' <image:tag>
• For services that do have arm64 images, don’t force amd64—running natively is faster.

Hypotheses (to help triage)

• UI expects a field that’s undefined when the API returns an empty or error payload (e.g., results.map(...) on undefined). A null/empty-state guard may be missing.
• Version mismatch: UI workbench-ui:1.0.6, services trustgraph-flow:1.2.20, CLI 1.3.9. If the API response schema changed between these, the UI could be mapping a field that the gateway/service no longer returns (or returns under a different key).
• Graph not populated: if KG extraction completed but storage/query path isn’t wiring correctly, tg-show-graph would be empty and UI vector search could receive an unexpected/empty shape from the gateway.

What would help me move forward

• Confirm the expected API response shape for Vector Search / Assistant endpoints that the UI is mapping over (field names/types).
• Confirm compatible versions matrix between workbench-ui, trustgraph-flow, and trustgraph-cli.
• Guidance on verifying KG population (which tables/collections to inspect for triples/embeddings, and which service logs to tail during a search).

[cybermaggedon]

TypeError: Cannot read properties of undefined (reading 'map')

That's a workbench error (which shouldn't be an error), the scenario is where there's nothign to search. So at least a workaround is to load some data in.

I think the request must be coming back null, rather than an empty array when there are no hits.

[cybermaggedon]

This error happens when a vector embeddings search is done on a non-existent collection.

That should be changed to not be an error. It's possible that this was happening because collection support in workbench was incomplete <= 1.1.3, you could be searching for a valid collection, and getting directed to 'default'.

These fixes are in the workbench, 1.1.4:
trustgraph-ai/workbench-ui#85
trustgraph-ai/workbench-ui#86
trustgraph-ai/workbench-ui#87

[cybermaggedon]

Hey @hdotking-materra there have been a few changes, trustgraph 1.3.19 + workbench 1.1.6 have changes to make this error message go away.

In answer to your question about the shape of the response, this error relates to the graph-embeddings query part of the API, which when given a vector returns a list of entities. With the above changes, the entities field will always be present in the response (unless there's an error) - it may be an empty list for a collection with nothing loaded.

[cybermaggedon]

Confirm the expected API response shape for Vector Search / Assistant endpoints that the UI is mapping over (field names/types).

See this brief summary of a search flow...

https://gist.github.com/cybermaggedon/89a72f80fb9b53410eda5c14875631ae

Confirm compatible versions matrix between workbench-ui, trustgraph-flow, and trustgraph-cli.

Currently, workbench-ui >= 1.1.3 is compatible with trustgraph-flow == 1.3.19 and trustgraph-cli == 1.3.19

• To be sure that versions are compatible, use a stable release from https://config-ui.demo.trustgraph.ai. We always publish configurations tested for compatibility.
• Once a release is stable, the trustgraph-* versions will be compatible, we don't publish incompatible changes to a release branch once the release is stable. So it is safe to upgrade (but not downgrade to risk pulling in an incomptable version). So e.g. you won't necessarily upgrade trustgraph-cli when upgrading other packages / containers.
• Pre-release products are likely to be exposed to major changes as this is development phase. May be useful for experimenting, but otherwise avoid.

Guidance on verifying KG population (which tables/collections to inspect for triples/embeddings, and which service logs to tail during a search).

For verifying KG load use tg-show-graph, specifying the flow ID and collection

usage: tg-show-graph [-h] [-u API_URL] [-f FLOW_ID] [-U USER] [-C COLLECTION]

Connects to the graph query service and dumps all graph edges.

options:
  -h, --help            show this help message and exit
  -u API_URL, --api-url API_URL
                        API URL (default: http://localhost:8088/)
  -f FLOW_ID, --flow-id FLOW_ID
                        Flow ID (default: default)
  -U USER, --user USER  User ID (default: trustgraph)
  -C COLLECTION, --collection COLLECTION
                        Collection ID (default: default)

There isn't an equivalent for the vector store currently. If you're using Qdrant port 6333 has a simple UI where you can see what collections are present, and see what's loaded.

http://localhost:6333/dashboard