> ## Documentation Index > Fetch the complete documentation index at: https://wavefront.rootflo.ai/llms.txt > Use this file to discover all available pages before exploring further. # Quickstart > Add a model, build an agent and a workflow, and run it from the Wavefront Console and the API In about fifteen minutes you will build a small **invoice checker**: 1. an **agent** that reads an invoice and returns its fields as JSON, 2. a **function** that checks the line items add up to the total, 3. a **workflow** that runs the two in order, and run it from the Wavefront Console and with `curl`. You need access to an app in the Wavefront Console, and an API key for one of the supported model providers. ## 1. Add a model Open your app and go to **LLM Repository → Add LLM**. | Field | Example | | - | - | | **Display Name** | `Default chat model` | | **Inference Engine Type** | Your provider, for example *OpenAI GPT* or *Azure OpenAI* | | **Model Type** | LLM | | **LLM Model Name** | The model or deployment name at your provider | | **API Key** | Your provider key | | **Base URL**, **API Version** | Only for providers that need them (Azure OpenAI, Ollama, vLLM) | Click **Add Model**, then copy the new entry's **id** from the list. Agents refer to the model by this id, so the key never appears in your YAML. ![The Add LLM to Repository dialog](https://mintcdn.com/rootfloai/Uz8lgLRL_3RYacDA/images/wavefront/qs-llm-config.png?fit=max\&auto=format\&n=Uz8lgLRL_3RYacDA\&q=85\&s=37caf2737e7966ef75511ff32a738ec3) ## 2. Create the agent Go to **Agents → Create Agent**. Enter an **Agent ID** (`invoice-reader`), pick a **Namespace**, and paste this YAML, replacing ``: ```yaml invoice-reader.yaml theme={null} metadata: name: invoice-reader version: 1.0.0 description: Reads an invoice and returns its key fields agent: name: invoice-reader role: Invoice reader job: | You read invoices and return their key fields exactly as printed. Use null for a field that is not on the invoice. Amounts are numbers without currency symbols or thousands separators. model: provider: rootflo # use a model from the LLM Repository… model_id: # …by its id settings: temperature: 0 parser: # the shape of the answer name: invoice-fields fields: - name: vendor type: str description: Name of the company that issued the invoice - name: invoice_number type: str description: The invoice number - name: invoice_date type: str description: Invoice date as YYYY-MM-DD - name: total_amount type: float description: The invoice total - name: line_items type: array description: Each line of the invoice items: name: line_item type: object description: One line fields: - name: description type: str description: What was charged - name: amount type: float description: Amount for this line ``` The `parser` makes the agent return JSON with exactly these fields. Field descriptions matter: the model reads them. ### Try it Open the agent. The console shows its name in title case (`invoice-reader` becomes *Invoice Reader*). In the **Ask anything** box, paste an invoice and send it: ```text theme={null} ACME Office Supplies Ltd. Invoice INV-2041, dated 3 March 2026. 1) Printer paper, 10 reams: 2,400.00 2) Toner cartridge: 3,150.00 3) Delivery: 250.00. Total due: 5,800.00 ``` The reply is structured: ```json theme={null} { "vendor": "ACME Office Supplies Ltd.", "invoice_number": "INV-2041", "invoice_date": "2026-03-03", "total_amount": 5800, "line_items": [ { "description": "Printer paper, 10 reams", "amount": 2400 }, { "description": "Toner cartridge", "amount": 3150 }, { "description": "Delivery", "amount": 250 } ] } ``` You can also attach a photo or a PDF of an invoice instead of pasting text: use **+ → Select Images** or **Select Documents**. ![Testing the invoice-reader agent in its chat panel](https://mintcdn.com/rootfloai/Uz8lgLRL_3RYacDA/images/wavefront/qs-agent-chat.png?fit=max\&auto=format\&n=Uz8lgLRL_3RYacDA\&q=85\&s=f4b033d9f26a190cd3a8288039c4da12) ## 3. Create the function The agent reads the invoice; a **function** does the arithmetic. Arithmetic, comparisons and decisions belong in functions: they are exact, testable and cost nothing to run. Go to **Functions → Create Function**. Name it `check-invoice-total`, paste the YAML below into the editor, and click **Create Processor**: ```yaml check-invoice-total.yaml theme={null} type: javascript description: >- Checks that an invoice's line items add up to its total. function: code: | export default function (input) { const items = Array.isArray(input.line_items) ? input.line_items : []; const sum = items.reduce((s, item) => s + Number(item.amount || 0), 0); const total = Number(input.total_amount || 0); const difference = Math.round((total - sum) * 100) / 100; return { invoice_number: input.invoice_number || null, vendor: input.vendor || null, total_amount: total, line_items_sum: Math.round(sum * 100) / 100, difference, status: Math.abs(difference) < 0.01 ? 'TOTAL_MATCHES' : 'TOTAL_MISMATCH', }; } input_schema: required: - total_amount properties: total_amount: type: number description: Invoice total, from the invoice reader line_items: type: array description: Line items, from the invoice reader ``` Test it from the function page with a sample input, for example the JSON reply from step 2. Copy the function's **id**. ## 4. Create the workflow Go to **Workflows → Create Workflow**. Enter a **Workflow ID** (`invoice-check`), pick the same namespace, and paste this YAML, replacing `` and ``: ```yaml invoice-check.yaml theme={null} metadata: name: invoice-check version: 1.0.0 description: Reads an invoice and checks that its line items add up to the total arium: agents: - name: invoice_reader role: Invoice reader job: | You read invoices and return their key fields exactly as printed. Use null for a field that is not on the invoice. Amounts are numbers without currency symbols or thousands separators. model: provider: rootflo model_id: settings: temperature: 0 parser: name: invoice-fields fields: - name: vendor type: str description: Name of the company that issued the invoice - name: invoice_number type: str description: The invoice number - name: invoice_date type: str description: Invoice date as YYYY-MM-DD - name: total_amount type: float description: The invoice total - name: line_items type: array description: Each line of the invoice items: name: line_item type: object description: One line fields: - name: description type: str description: What was charged - name: amount type: float description: Amount for this line function_nodes: - name: check_total function_name: trigger_message_processor # runs one of your functions description: Checks the line items against the total input_filter: [invoice_reader] # receives the agent's answer prefilled_params: message_processor_id: workflow: start: invoice_reader edges: - from: invoice_reader to: [check_total] end: [check_total] ``` How to read it: * `agents` and `function_nodes` declare the steps; `workflow` wires them together. * `input_filter: [invoice_reader]` hands the agent's JSON to the function. The function sees its fields directly: `input.total_amount`, `input.line_items`. * The workflow's output is the output of its last node, `check_total`. ### Run it in the console Open the workflow. Its YAML is on the left; on the right is the same chat panel as for agents. Paste the invoice text from step 2 into **Ask anything** and send. (This workflow needs no variables; when one does, add them with **+ → Add variables**.) The result: ```json theme={null} { "invoice_number": "INV-2041", "vendor": "ACME Office Supplies Ltd.", "total_amount": 5800, "line_items_sum": 5800, "difference": 0, "status": "TOTAL_MATCHES" } ``` Change a line amount in the input and run again to see `TOTAL_MISMATCH`. ![Running the invoice-check workflow in the console](https://mintcdn.com/rootfloai/Uz8lgLRL_3RYacDA/images/wavefront/qs-workflow-run.png?fit=max\&auto=format\&n=Uz8lgLRL_3RYacDA\&q=85\&s=108598e8cdcde437800635389508b183) ## 5. Run it from the API Your application calls the app directly, with an **app user** (not your console account). Sign in to get a token: ```bash theme={null} APP=https://your-app.example.com TOKEN=$(curl -s -X POST "$APP/floware/v1/authenticate" \ -H 'Content-Type: application/json' \ -d '{"email": "builder@example.com", "password": "…"}' \ | jq -r '.data.user.access_token') ``` Then run the workflow by its id (shown on the workflow page): ```bash theme={null} curl -s -X POST "$APP/floware/v2/workflows//inference" \ -H "Authorization: Bearer $TOKEN" \ -H 'Content-Type: application/json' \ -d '{ "inputs": [{ "role": "user", "content": "ACME Office Supplies Ltd. Invoice INV-2041 … Total due: 5,800.00" }], "variables": {} }' ``` ```json Response theme={null} { "meta": { "status": "success", "code": 1, "error": null }, "data": { "message": "Workflow inference completed successfully", "data": { "result": "{\"difference\": 0, \"invoice_number\": \"INV-2041\", … \"status\": \"TOTAL_MATCHES\"}", "workflow_id": "invoice-check", "execution_time": 2.38, "variables": {} } } } ``` > [!NOTE] > > A workflow's `result` is a **string** containing the last node's output; parse it as JSON. An agent run through `POST /floware/v2/agents//inference` returns `result` as an object when the agent has a parser. Every response uses the same envelope: `meta.status` is `success` or `failure`, and on failure `meta.error` says why. > [!WARNING] > > Signing in again with the same app user ends that user's other sessions. Give each script or service its own user. ## What you built You used the pattern most Wavefront features follow: **an agent gathers the facts, a function makes the call, a workflow runs them in order.** ## Next steps > **[Workflows in depth](/wavefront/workflows)** > > Routers, sub-workflows, iterators and versions. > **[Read and write your data](/wavefront/datasources)** > > Load the record to check from a database and save the result. > **[Rules as configuration](/wavefront/configurations)** > > Move the tolerance into a configuration admins can edit. > **[Run in the background](/wavefront/running-workflows)** > > Queue long runs and poll for their result. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.