Skip to content

SETUP.md

Latest commit

 

History

History
152 lines (113 loc) · 3.33 KB

SETUP.md

File metadata and controls

152 lines (113 loc) · 3.33 KB
Raw

Langflow Setup For Assembly Optimizer

Langflow is an optional visual demo layer. The backend remains the source of truth for physical calculation, Harness validation, policy gating, HITL, and dispatch artifacts.

Recommended Import

Use:

langflow_integration/assembly_optimizer_multiagent_import_ready_flow.json

This import-ready flow is wired for:

  • POST /analyze
  • POST /multiagent/runs
  • POST /multiagent/runs/{thread_id}/resume
  • GET /multiagent/runs/{thread_id}

Optional Custom Component

The source component is:

langflow_integration/assembly_optimizer_component.py

Paste it into Langflow custom components, or place it in the directory your Langflow instance uses for local components.

Start Backend

powershell -ExecutionPolicy Bypass -File .\start_stack.ps1

Default backend:

http://127.0.0.1:8010

Supported Operations

  • analyze
    • Calls POST /analyze
    • Use for deterministic press-fit analysis.
  • multiagent_run
    • Calls POST /multiagent/runs
    • Starts LangGraph orchestration and may return a HITL interrupt package.
  • multiagent_resume
    • Calls POST /multiagent/runs/{thread_id}/resume
    • Use after a reviewer approves or rejects a paused run.
  • multiagent_state
    • Calls GET /multiagent/runs/{thread_id}
    • Reads the latest stored graph state.

Demo Sequence

  1. Start the backend.
  2. Import assembly_optimizer_multiagent_import_ready_flow.json.
  3. Set Operation = multiagent_run.
  4. Paste a higher-risk payload such as tests/golden/case_002_thermal.json.
  5. Run the flow.
  6. If the output status is waiting_for_approval, copy thread_id.
  7. Set Operation = multiagent_resume.
  8. Paste the thread_id.
  9. Set Approval Decision = approve or reject.
  10. Run again to continue from the checkpoint.

Optional LLM Settings

The gateway exposes these fields:

  • LLM Enabled
  • LLM Provider
  • LLM Model
  • LLM API Base URL
  • LLM API Key
  • LLM Temperature
  • LLM Max Tokens

Leave LLM disabled for deterministic offline demos. Use provider = mock for offline specialist enrichment.

Example config:

{
  "llm": {
    "enabled": true,
    "provider": "mock",
    "model": "industrial-mock"
  }
}

For a live OpenAI-compatible endpoint, provide your own local secret. Do not commit keys:

{
  "llm": {
    "enabled": true,
    "provider": "openai_compatible",
    "model": "<model-name>",
    "api_base_url": "https://api.openai.com/v1",
    "api_key": "<provider-api-key>",
    "temperature": 0.0,
    "max_tokens": 700
  }
}

Gateway Headers

If you put the backend behind a local proxy, auth gateway, or cloud tunnel, fill:

  • API Key
  • API Key Header
  • Headers JSON

Examples:

{"X-API-Key": "local-demo-key"}

or:

API Key Header = Authorization
API Key = <gateway-token>

The component will send:

Authorization: Bearer <gateway-token>

What Langflow Should Do

  • Collect or transform user-friendly input.
  • Call the backend.
  • Narrate engineering results.
  • Present recommendation, risk, report paths, and HITL package.
  • Let a reviewer decide whether to resume a paused run.

What Langflow Should Not Do

  • Recompute contact pressure or safety factor.
  • Override Harness failures with free-form text.
  • Bypass the backend policy gate.
  • Dispatch anything directly to equipment.