Skip to main content

Troubleshooting

Use this page when the local GTM showcase does not behave as expected.

Agent returns 401 from /anip/tokens​

Cause:

  • the service stack is missing demo actor API keys,
  • the agent runtime is pointing at the wrong service port,
  • the selected language compose stack was not recreated after image changes.

Check:

docker compose -p anip-gtm-go-local -f examples/showcase/gtm/docker-compose.language-parity-go.yml ps

Then hit the token endpoint directly with the demo actor key configured in the compose file.

Metabase shows setup page​

Cause:

  • gtm-metabase-setup did not complete,
  • Metabase data volume belongs to an older failed setup,
  • the wrong Metabase port is open.

Rerun:

GTM_METABASE_URL=http://127.0.0.1:3043 \
  python3 examples/showcase/gtm/scripts/setup_metabase_verification.py

Open pipeline values are blank or not available​

Correct open pipeline logic uses:

coalesce(close_value, sales_price)

If blank values appear, check:

  • mart_gtm__pipeline_health.sql
  • bi_gtm__forecast_stage_summary.sql
  • bi_gtm__product_pipeline.sql
  • generated/custom backend adapters for the language under test

Then rebuild dbt and recreate the service containers.

Agent chooses the wrong capability​

Check:

  • capability brief generated by live discovery,
  • agent-consumption metadata,
  • whether the question is missing required inputs,
  • whether a composed capability should ask for clarification instead of guessing.

The fix should usually be in contract metadata, source evidence, or custom bundle behavior. Avoid hardcoded phrase patches in generic runtime code.

Question bank differs across languages​

Check:

  • same package version,
  • same custom bundle generation,
  • same service ports,
  • same model and base URL,
  • same approval store state,
  • same dbt data state.

Language-specific bugs belong in that language bundle or runtime target. Contract drift belongs in Studio/package generation.