What you can build

The Vertical AI platform exposes two kinds of endpoints:

  • Conversational gateways that wrap an LLM around a tool catalog. Send a message, get back a stream of tool calls, intermediate results, and a final assistant response. One gateway per agent (overhead, underground).
  • Engineering services that do one thing well — parse a SPIDA file, run a clearance check, return the loading rules for a pole, render a JP Form 2 PDF. Use these when you want a single piece of the pipeline without orchestration.

Both are HTTP-and-JSON. Both authenticate with the same bearer token.

What you need

  1. An API key. Email hello@epcstudio.io for a sandbox key.
  2. A base URL. The conversational gateways live at:
    • Overhead: https://hv-pole-platform-dev.web.app/api
    • Underground: https://hv-pole-platform-dev.web.app/api/ug
  3. A pole or vault model, or a source file we can parse into one (SPIDA, O-Calc, PLS, drone and field data for overhead; vault/conduit JSON for underground).

A 30-second example

Open a chat session against the overhead gateway and ask for a summary of an uploaded pole:

curl -N https://hv-pole-platform-dev.web.app/api/chat \
  -H "Authorization: Bearer $VERTICAL_AI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "session_id": "demo-1",
    "message": "Parse my SPIDA file and tell me what is on pole 3."
  }'

The response is a Server-Sent Event stream. You’ll see events for tool calls (run_circuit_classification, summarize_pole, etc.), tool results, and finally the assistant’s text reply.

What’s next