Building your first agent in Avery NXR - a 10-minute walkthrough

If you've installed Avery NXR and you're looking at the Agents tab wondering where to start, this walkthrough is for you. We'll build a working agent in about 10 minutes, end to end, using nothing but the templates and connectors that ship with the product.

The example workflow: an agent that watches your email for invoices, extracts the line items, and posts them to your accounts payable queue with the vendor and amount already classified. By the end of this, you'll have it running on real data.

Step 1 — Open the Agents tab (15 seconds)

Launch Avery NXR. In the top nav, click Agents. You'll see two views: the running agents (empty if you haven't built any yet) and the templates panel.

We'll start from a template because it's faster and the templates are how most real agents get built — pick one that's close to what you want, then customize.

Step 2 — Pick the Invoice Processor template (30 seconds)

Click + New Agent and choose the Template option (vs. Blank or Wizard).

In the template grid, find the Invoice Processor. This is one of Avery's pre-loaded production templates. It's set up to:

• Watch an email inbox for invoices
• Extract structured fields (vendor, amount, line items, due date)
• Classify the type (utility, expense, service, etc.)
• Match against open POs
• Route to approval if over a threshold
• Post to accounts payable

Click Use this template. Avery opens the agent's graph in the visual builder.

Step 3 — Connect your email (2 minutes)

The first node in the graph is the Email Trigger. By default it watches the ap@ inbox. You need to connect a real email account.

Click the Email Trigger node. In the right-hand panel, click Connect inbox.

You'll see two options:

• IMAP — for any email provider with IMAP enabled
• OAuth — for Gmail or Microsoft 365

Pick one. If OAuth, you'll go through the standard Google or Microsoft consent flow. The credentials are stored in your OS keychain, not in Avery's cloud (because there isn't one).

Once connected, set the inbox or label you want monitored. For testing, point it at a Gmail label you can drop test emails into.

Step 4 — Verify the local model is running (30 seconds)

In the bottom-right of the window, you should see a model indicator — something like qwen2.5-coder-7b · ready.

If you don't see a ready state, click into Settings → Local model and pick a model. Avery will rank the available models against your hardware. Pick the recommended one if you're unsure.

The model only needs to be loaded once. Subsequent agent runs hit the warm model.

Step 5 — Walk through the rest of the graph (3 minutes)

The Invoice Processor template has 12 nodes. Walk through them in order:

• Email Trigger — already configured
• Extract — pulls vendor name and amount from the email body or attached PDF. This calls the local LLM.
• Classify — labels the invoice (utility / expense / service). Local LLM.
• Match PO — does a vector search against your open POs (if you've configured a PO data source). Vector KB.
• Approve? — a conditional. If amount > $1,000, route to approval.
• Post to GL — calls your accounting integration (NetSuite, QuickBooks, Xero, or HTTP webhook to your custom system)
• Notify AP — posts to Slack with the structured summary

For each node, click into it to see what it does and what configuration it needs. Most of the nodes work out of the box — you mostly need to wire up the integrations (the GL system and Slack).

Step 6 — Wire up Slack (1 minute)

Click the Notify AP Slack node. If you haven't connected Slack yet, you'll get the OAuth flow. Authorize. Pick the channel where invoice notifications should land (e.g., #ap-review).

The connector is stored once and reusable across every agent and app in Avery NXR. You won't need to reconnect Slack for the next agent you build.

Step 7 — Test on a real invoice (2 minutes)

Drop a real invoice PDF into the email inbox you connected, or send one to it.

In the Avery window, you should see the agent trigger within a few seconds. The graph nodes will light up one by one as they execute. You can click into each node to see the input and output it produced.

If something doesn't look right, click into the audit log (the Audit tab on the project) to see the full step-by-step record of what the agent did and why.

When everything looks right, click Activate. The agent is now live.

Step 8 — Adjust thresholds and rules (1 minute)

Once you've seen a few invoices process, you'll know what to adjust. Common edits:

• Approval threshold — change the $1,000 default to fit your policy
• PO matching confidence — tighten or loosen based on accuracy
• Notification channel — route urgent invoices to a different channel
• Auto-post threshold — automatically post low-risk invoices, escalate the rest

Every change is a click on a node and a property update. No code edits required.

What just happened

In about 10 minutes, you wired up an agent that:

• Watches an email inbox
• Reads incoming invoices (handles PDF parsing)
• Extracts structured fields using your local model
• Classifies and routes based on rules and confidence
• Connects to your accounting system
• Notifies your team

All running on your laptop. Every action audited. Every node inspectable. Every credential stored in your OS keychain.

Where to go next

A few obvious extensions once the basic agent is working:

• Add a PO data source so the Match PO node has something to match against
• Add a vendor master so Avery can learn your vendor naming conventions
• Fork the agent for a different workflow — expense receipts instead of invoices, contracts instead of invoices, customer signed forms

The visual builder makes forking trivial. Right-click → Duplicate → modify the trigger and the extraction logic. You have a new agent.