Skip to content

Latest commit

 

History

History
195 lines (132 loc) · 4.7 KB

File metadata and controls

195 lines (132 loc) · 4.7 KB

Choose A Blueprint Example

This guide helps you choose a checked-in OtterDesk blueprint for a first run, smoke test, or runtime demonstration.

Run commands from the workspace root unless a step says otherwise.

List Available Blueprints

mn blueprint list

Expected output includes:

ID
Name
Job Name

Catalog availability depends on your cached blueprint repository. The checked-in local catalog for this workspace is otterdesk-blueprints/index.json.

Recommended Order

Order Blueprint Use it when
1 tax_form_ocr_capture_assistant You want a small document/OCR workflow with public sample metadata.
2 portfolio_risk_review_assistant You want a finance workflow with deterministic simulation and LLM report writing.
3 video_watch_assistant You want a service-style video monitoring workflow.
4 gtm_ai_workflow You want a long-running GTM automation loop with CRM and outreach artifacts.
5 personal_financial_advisor You want a folder-watching service with OCR, public browser research, and review-only reports.
6 generic_customer_service_voice_coworker You want a local voice service backed by the NVIDIA/Spark stack.

1. Tax Form OCR Capture Assistant

Path:

otterdesk-blueprints/tax_form_ocr_capture_assistant

Use it when:

  • you want a compact checked-in blueprint;
  • you want document intake and OCR-style artifact output;
  • you want a review-only workflow with public dataset notes.

Run:

mn blueprint run --folder otterdesk-blueprints/tax_form_ocr_capture_assistant

Direct runner smoke test:

cd otterdesk-blueprints/tax_form_ocr_capture_assistant
python3.11 payloads/document_workflow/scripts/run_blueprint.py --runs-root /tmp/mn-runs --run-id tax-form-demo

2. Portfolio Risk Review Assistant

Path:

otterdesk-blueprints/portfolio_risk_review_assistant

Use it when:

  • you want public market data plus deterministic risk simulation;
  • you want the LLM limited to interpretation and report writing;
  • you want a clear review-only human approval policy.

Run from the catalog:

mn blueprint run portfolio_risk_review_assistant

Run from the local folder:

mn blueprint run --folder otterdesk-blueprints/portfolio_risk_review_assistant

3. Video Watch Assistant

Path:

otterdesk-blueprints/video_watch_assistant

Use it when:

  • you want a service blueprint;
  • you want visual detection events and cooldown state;
  • you want to inspect live run status and service artifacts.

Run:

mn blueprint run --folder otterdesk-blueprints/video_watch_assistant --web-ui

Cancel the service when done:

mn job cancel <job_id>

4. GTM AI Workflow

Path:

otterdesk-blueprints/gtm_ai_workflow

Use it when:

  • you want a long-running account research and outreach loop;
  • you want local CSV CRM and market-insight artifacts;
  • you want to test email/inbox integrations in dry-run or test-recipient mode first.

Run:

mn blueprint run --folder otterdesk-blueprints/gtm_ai_workflow

Inspect run state:

mn blueprint monitor --follow

5. Personal Financial Advisor

Path:

otterdesk-blueprints/personal_financial_advisor

Use it when:

  • you want a continuous local folder-watch service;
  • you want OCR plus public, privacy-safe browser research;
  • you want review-only household finance reports and risk reminders.

Bounded direct runner smoke test:

cd otterdesk-blueprints/personal_financial_advisor
python3.11 payloads/document_workflow/scripts/run_blueprint.py --runs-root /tmp/mn-runs --run-id personal-finance-demo --watch --max-cycles 1

6. Pizza Order Voice AI Co-worker

Path:

otterdesk-blueprints/generic_customer_service_voice_coworker

Use it when:

  • you want a real-time voice service;
  • you have the local NVIDIA/Spark stack ready;
  • you want editable menu knowledge and a localhost HTTPS voice page.

Run with longer cold-start timeouts when the NVIDIA stack needs to warm up:

MN_PRE_LAUNCH_TIMEOUT_SECONDS=900 NEMOTRON_PRELAUNCH_WAIT_SECONDS=900 mn blueprint run generic_customer_service_voice_coworker

Security Notes

  • Review manifest.json before running any blueprint.
  • Check pass_env before passing secrets.
  • Use mock, sample, quick-test, and dry-run modes before external delivery.
  • Treat financial, tax, legal, healthcare, and safety outputs as review-only until a qualified human approves them.
  • Cancel service jobs when you are done.

Related Pages