Building on the layer

An operations manager routes a sentinel digest to Slack. A coordinator builds a workflow for engineer follow-ups. A developer opens Cursor and asks it to build an account review panel against the VH3 AI API. All three read the same customers, jobs, sites, and engineers. In this guide, the substrate names the decoupled, persistent operational knowledge graph and vector index in the customer’s account. VH3 AI maintains that substrate independently of model inference.

• The ingestion pipeline enriches jobs and relationships once, then stores them for the organisation.
• Deterministic search, similarity, entity resolution, and customer knowledge sections read the substrate without an LLM.
• Agents, reports, and automations read the same records when they need language.
• Email, calendar, CRM, storage, and field systems add people, jobs, and sites to the same graph through managed sync.

The substrate persists across sessions. An LLM context window and its generated response support ephemeral inference for one request. Builders can replace or reroute the model while the account keeps its graph, vector index, enriched records, and operating rules. This guide covers builder workflows for operations staff using no-code tools and technical staff using Claude Code, Cursor, or similar coding agents.

​

What integrated AI looks like in field service

VH3 AI supports levels 3 and 4. Level 2 makes each session reread raw text and leaves the organisation without a structured record it can keep.

​

Three builder paths, one substrate

All three paths call the same APIs and read the same operational model. A job remains a job, and a customer remains a customer, because VH3 AI resolves every incoming record into one field service domain model before it enters the graph. VH3 AI keeps the detailed schema private. Every endpoint returns consistent, resolved, connected entities. An app or automation on this layer inherits that contract. See The domain model in the intelligence layer guide.

​

What an app or workflow can call

Connie uses these tools internally. Your agents, automations, and apps can call the same capabilities. Sentinels notice change. Discovery and synthesis answer questions. Cases and teams hold ownership and follow-through. The full inventory and evaluation notes for AI agents live in Platform tools.

​

Connected tools and programmatic access

Field service already spans several systems. VH3 AI brings those records into one operational model so mail, calendars, CRM, storage, and field data attach to the right customer and job.

​

Native connections in the platform

Native integrations run inside VH3 AI. VH3 AI handles OAuth, manages sync, and shows connection health to admins. Typical categories include: Per-user connections. An engineer’s inbox and calendar can inform briefings within the scope that the engineer authorised. The connection does not expose everyone’s mail to the organisation. Organisation-level connections. A Xero or Slack workspace and a CRM can supply context to automations and reports across the team.

​

Programmatic access for agents and apps

The platform exposes its internal capabilities through APIs.

• REST API for search, jobs feeds, sentinels, reports, Connie, cases, teams, and backfill tasks.
• MCP server so Claude Desktop, Cursor, and other MCP clients can call the tools Connie uses. The server handles credentials after JWT authentication.
• n8n community node for workflow builders who want automation without writing a backend.

When email arrives or a calendar event appears, ingestion and entity resolution map people, domains, and addresses back to contacts in the model. Personal agents and shared automations then use the same identity.

​

Operations builders, no-code and low-code

Field service teams can ship local fixes themselves.

• An operations manager routes a sentinel digest to Slack when SLA performance slips.
• A contracts lead schedules an account report to a client distribution list each month.
• A service coordinator creates a workflow when engineer-flagged follow-ups spike for one customer.

Templates, connectors, and a scoped API give each workflow the same data contract. Starting points

• n8n node and templates
• Agent kits for n8n agents
• Claude Projects for ops

​

Coding agents on a secure substrate

Claude Code and Cursor need a clear domain contract. Give them:

1. A stable API contract. Use OpenAPI and consistent field names.
2. Guardrails. Define which endpoint answers each question and which fields end users must never see.
3. A tenant boundary. Send company_id and api_key on the server and prevent cross-customer leakage.

VH3 AI ships Agent Starter Kits with AGENTS.md, Cursor rules, MCP setup, and n8n prompts. These assets encode field service routing across discovery, synthesis, and sentinels.

​

What coding agents should build

​

Safe patterns

​

MCP without writing a backend

The MCP server exposes search, investigate, sentinels, jobs feeds, and reports. A coding agent can call operational discovery directly, then generate UI or workflow code from the response. See MCP setup.

​

Customer knowledge your agents can fetch

Builders should understand the Customer Summary object. See Operational discovery.

• Seven modular sections support independent search and ranking.
• The platform refreshes sections on a schedule or when job drift thresholds change, so the brief stays current as jobs arrive.
• Connie receives the sections with recent jobs since generation, so each conversation starts with current context.

Coding agents can call POST /search/summary-sections for thematic queries across accounts, or fetch a full brief per contact before rendering a custom account page. For a parent contact with multiple child sites, such as a retail chain or housing association, the Account Summary combines each child Customer Summary into one account-level brief and refreshes it on the same schedule. Call GET /search/account-summary/by-parent/{company_id}/{contact_id} to fetch it, or use POST /search/account-summary-sections for account-level search. Pass any child contact ID and the lookup resolves to the parent automatically. See Operational discovery account summary for the full shape.

​

Turn a sentinel into a case

VH3 AI pairs detection with ownership so a signal can become accountable work. Example flow

1. A sentinel flags repeat attendance for a contact.
2. An automation runs search/outcomes for similar faults on that account.
3. The workflow opens a case called “Third fire panel callout in six weeks” and links the relevant jobs.
4. The regional team receives a notification, and Connie drafts an engineer briefing for the next visit.

Cases and teams stay lightweight. Status, participants, linked jobs, and ownership give the team enough structure to finish the work.

• Cases API
• Teams API

​

Why coding agents work here

A coding agent that rediscovers customers on every click burns tokens and can still choose the wrong name. The ingestion pipeline already enriches records, resolves multiple entities, and links jobs, customers, sites, and history. Agents can then generate thin applications that call well-shaped endpoints.

Use the API to build operational tools that fit your people and their work.

Your FMS still creates and closes jobs. Your IDE still writes the code. VH3 provides the substrate that both systems use.

​

Security builders actually hit

Builders should assume:

• Every call carries company_id and a validated api_key. The API enforces organisation boundaries.
• Built-in user management and auth support invites, roles, and JWTs.
• PII handling follows your DPA. Do not rebuild sensitive fields in public UIs.
• Field integration stays read-only by default. Enable write-back only after explicit agreement.

See Authentication for the credential map and Deploying secure apps for the deployment checklist.

​

Choose your starting kit

​

Related

​

Next step

Create a server-side route for one test tenant. Call GET /search/outcomes for a known customer, check the returned job references against the source record, and only then add a sentinel or model-backed step to the workflow.