How Top AI Apps Fit Together for Solo Owners and Small ...

A contractor doesn't need another place to lose the estimate. A shop owner doesn't need product photographs trapped in a chat. The useful integration ends with a checked file or an updated record in the place the business already uses.

a16z's seventh Top 100 Gen AI Consumer Apps report includes the ingredients: Notion, Gamma, Fireflies, Photoroom, ElevenLabs, OpenRouter and n8n. Our companion hub reads the rankings and explains the control tradeoffs. This article follows the work between those apps.

These are proposed workflows, not customer case studies. The connections below are documented by their vendors as of October 16, 2026. I haven't run these chains against a business's accounts. Start with a sample that contains no customer information and check each handoff before connecting real work.

Pick the connection that fits the job

A native node gives n8n a named operation. Its Notion node can create and update database pages and append blocks. Its Gmail Draft operation can create an email draft, including attachments. Those two nodes can put an approved work summary into Notion and prepare the corresponding email. n8n is the bridge between them; this isn't a direct Notion-to-Gmail feature.

An API call is useful when the service's documented operation is the best fit. n8n's HTTP Request node supports methods, authentication, request bodies and file handling. For the examples below it calls Photoroom, ElevenLabs or Fireflies. These are configured API calls through n8n, not a claim that every one has a native node.

A webhook starts work when another system sends an event. n8n's Webhook node supplies a receiving address and supports authentication. It doesn't make an old booking system capable of sending events. If that system has no documented outbound connection, use an owner-initiated handoff instead. None of the chains here assumes an undocumented connection to a property, medical or accounting system.

A connector inside an assistant suits occasional, supervised work. Gamma documents a ChatGPT app and Claude connector that create Gamma content from a conversation. Both still create work in Gamma's service. Gamma says these connectors work on all its plans without an API key; the assistant's own access requirements are separate.

For repeatable work, the same page documents Gamma's official n8n node, with generation and status operations. Its API access guide lists Pro, Ultra, Teams and Business for API keys. Don't assume a connector entitlement also includes API access.

An exported file can be the entire integration. Gamma's export guide documents PDF, PNG and PowerPoint. Downloading a PDF and manually placing it in an existing job folder needs no connection to that folder's software. It does need someone to check that the export is complete.

The existing n8n guide covers setup. The MCP guide covers connector selection and permissions. Here, the choice is where a particular piece of work should go.

A one-person business, through a normal week

Consider an independent designer who sells a defined service. Keep the starting set small: Notion for the work queue, n8n for transfers, Gamma for proposals, Gmail for correspondence and Stripe for collecting payment. The owner's project folder holds approved originals and delivered files.

Capture on Monday. The owner enters a short project brief into Notion after reading the inquiry. Keep the customer address and private attachments in the existing project records. A manual n8n run reads the selected Notion database page. This avoids treating an incoming message as an instruction to the automation.
Draft on Tuesday. n8n sends the approved brief to Gamma using its native node. It checks generation status before treating the result as ready. For an API implementation, Gamma documents creating a generation and polling its status, including an optional exported file. The owner checks scope, fees and exclusions against their original notes.
Publish on Wednesday. The owner downloads the checked PDF into the project folder and manually uploads the public version to the business's existing website. That upload is a deliberate manual handoff, not an asserted Gamma connection to the site's publisher. Client details stay out of a public portfolio version.
Follow up on Thursday. From an owner-approved summary, n8n creates a Gmail draft with the proposal attached. The owner opens it, confirms the recipient and sends it. Later follow-ups start from the owner's review of the work queue, rather than a model guessing whether the customer has replied.
Get paid when agreed. The owner uses Stripe's dashboard to create an invoice for that customer and manually adds its link to the approved message. Stripe's comparison of Invoicing and Payment Links distinguishes a customer-specific invoice from a reusable link for a product. The customer pays on Stripe's hosted page. The owner checks payment in Stripe and marks the project paid; an AI draft doesn't establish that money arrived.

In this proposed setup, n8n and the project folder run on an office machine the owner maintains. Notion, Gamma, Gmail and Stripe are rented services. The brief reaches Notion and Gamma; correspondence and its attachment reach Gmail and the recipient; billing details reach Stripe. A local copy of the proposal doesn't remove those service copies.

If the owner wants unattended webhooks while the office machine is asleep, that requires an always-available receiver. Renting n8n hosting moves its workflow data and execution records to that host. Make that a conscious change. The workstation guide covers the operating choices without needing a second server tutorial here.

Set up first: selected Notion page to Gmail draft. Prove that the correct project, recipient and attachment survive the transfer. Add Gamma after that small path is dependable.

When a local model or OpenRouter belongs in the chain

An optional drafting stage can sit before Gamma or Gmail. n8n documents an Ollama Chat Model node for local models and an OpenRouter Chat Model node for hosted calls. They are model components used with a suitable n8n chain or agent, not standalone replacements for the entire workflow.

For a public service description, the owner could send approved wording through OpenRouter to a chosen provider and inspect the draft before using it. OpenRouter's bring-your-own-key documentation describes storing provider keys with OpenRouter and using them for routed requests. The n8n credential authenticates to OpenRouter; the provider key belongs in OpenRouter's BYOK settings. A ChatGPT or Claude app subscription is not that provider API key.

BYOK still sends the request through OpenRouter to the provider. It doesn't turn the hosted model into a local one. The documentation distinguishes prioritized and fallback keys and an option to use a key always; review the failure path as well as the preferred route. Existing provider routing coverage handles the broader design.

For private drafting, keep both the model endpoint and the workflow on the owner's machine, with outside calls disabled and checked. Even then, sending the resulting text to Gamma or Notion sends that text outside. Choose what crosses the boundary before adding the cloud stage.

Eight verticals, with the first chain worth building

Bookkeeping and tax preparation

The bookkeeper finishes checking an internal document checklist. That approval starts a local n8n run.
A local model drafts clearer instructions from approved document categories. Actual returns, account numbers and client identifiers stay in the accounting system and local working folder.
The bookkeeper checks the wording. Only the generic approved version goes through the Notion node into a reusable instructions page.
For a particular client, the bookkeeper manually selects the relevant instructions and recipient, then asks n8n to create a Gmail draft. They check it against the actual file and send it themselves.

The bridge is n8n's documented Notion and Gmail operations. No connection to the tax package is assumed. This adds a controlled publication step to the hub's local drafting idea without letting a model decide what the client owes or must file.

Set up first: approved generic checklist to Notion. Keep client-specific drafting out until the generic path is correct.

Self-storage and property management

The manager approves a changed public instruction, such as where delivery drivers should park. That starts an owner-initiated n8n run.
HTTP Request sends only that script to ElevenLabs' text-to-speech endpoint, which returns audio and supports an MP3 output format.
The manager listens to the recording and saves the approved audio with its script. They manually upload it to a phone system only after checking that system's accepted format. There is no claimed direct phone integration here.
A staff instructions page in Notion records the approved wording and review date through the Notion node. Tenant names, access codes and balances stay in the property system.

Set up first: one public instruction to a reviewed audio file. The hub covers voice rights and commercial use conditions; keep that check with the recording. This chain does not answer live tenant account questions.

Motels and small lodging

A manager approves a revision to a housekeeping procedure in Notion.
A manual n8n run reads that page and sends the approved training text to Gamma's native node.
The workflow checks status. The manager reviews the resulting deck and exports a PDF for the staff binder or existing shared folder, placing it there manually.
The manager records the accepted version back in Notion. Guest records, room access codes and occupancy details never enter this training chain.

Set up first: one approved procedure to one checked PDF. It builds a repeatable handoff around the hub's training example. Booking and maintenance forms follow the separate person-sends rule; a generated deck cannot change a reservation.

Restaurants and small caterers

The owner approves a catering package description, with checked prices and ingredient wording, in Notion.
n8n reads that entry and asks Gamma to arrange it as a document. Use the approved text as the source; review every condition again in the result.
The owner exports a PDF, saves the approved version with the menu originals and manually publishes it through the restaurant's existing website editor.
For a customer inquiry, the owner selects that package and uses n8n to prepare a Gmail draft with the checked PDF attached. The owner confirms availability and sends the quote.

Allergen decisions, substitutions and customer dietary information stay with the restaurant's staff and existing records. This chain doesn't connect to the point-of-sale system or promise live availability.

Set up first: approved package description to a reviewed PDF. It's a different task from the hub's music example and doesn't require another model subscription.

Home services and trades

After a permitted site conversation has been transcribed in Fireflies, the contractor selects its transcript ID and starts n8n manually.
HTTP Request calls the Fireflies transcript query. Its documented response fields include sentences and summary material. n8n retrieves only the fields needed for the job.
n8n appends the selected material to a Notion job page. The contractor compares it with their own measurements and notes, marking unresolved choices for review.
After approval, n8n creates a Gmail draft of the scope questions. The contractor checks and sends it; the accepted scope is manually filed in the existing job folder.

Fireflies already holds the recorded conversation, and this transfer creates another copy in Notion. Keep alarm codes and financial documents out of the recording; local review later cannot undo their earlier disclosure. This example assumes the account has API access, not that every plan does.

Set up first: selected transcript to a review page. Leave quotation amounts and contract acceptance under the contractor's control. This complements the hub's Otter export option with a separately verified API route.

Healthcare offices

An office manager edits a generic opening checklist on the office machine. It contains no patient facts.
A local model can suggest clearer wording. The manager checks the result against the actual procedure before approving any transfer.
n8n sends only the approved generic text to Gamma to arrange a staff training document.
The manager exports and checks a PDF, then manually places it in the existing staff training folder. Patient records and encounter notes stay in the clinical system and out of this chain.

Set up first: a nonclinical office checklist to a checked PDF. This is an administrative proposal, not a clinical recommendation or a claim that these accounts are approved for patient information. The hub covers the separate agreement and plan questions for recording patient encounters. OpenEvidence remains a clinician's reference in that discussion; no OpenEvidence-to-medical-record integration is asserted here.

Retail and e-commerce

The owner selects an actual product photograph and product ID in a local work queue. Starting the run is the trigger.
n8n's HTTP Request node uploads the image to Photoroom using the documented Image Editing API request. It accepts a file upload and a background removal parameter and returns an image. Photoroom labels this the Plus plan API; don't assume an app subscription grants the same access.
The owner downloads and checks the returned image beside the original, retaining the product ID in their own catalog. Buyer records and margins stay out of the upload.
The owner manually uploads the approved image to the existing shop editor. That last step is a manual handoff, not an asserted Photoroom connection to every storefront.

Set up first: one photograph to one inspected output. Check handles, transparent packaging and existing damage before publishing. The new part here is the file transfer around the hub's image idea, not an automatic listing writer.

Legal practices and independent consultants

The owner approves a public description of a service in a local document. Privileged material and client matter files stay in the practice's existing system.
In a supervised Claude conversation, the owner supplies only that approved public text and uses Gamma's documented connector to create a service briefing.
They review the briefing in Gamma, export a PDF and manually place it with the public website assets.
They can record the accepted version in Notion. An optional assistant connection uses Notion's hosted MCP server, which documents reading and updating authorized content. That is a separate connection, not a direct Gamma-to-Notion transfer.

Notion's server is hosted by Notion, even when the client runs on the office machine. Granting a connector access to a workspace is a data decision; use a workspace containing only the material intended for that connection.

Set up first: approved public service text to a reviewed briefing. This adds a vertical missing from the hub while avoiding a second guide to legal document drafting or practice management.

Make the handoff recoverable

Keep a record of the source version, destination, result and owner's approval. Before retrying a failed generation, check its status; before creating a second page or email draft, look for the first one. A repeated run should not quietly create duplicate work.

Gamma's status documentation warns that its export link allows anyone holding it to download the file until it expires. Download the approved file into the intended private folder rather than placing that temporary link into a public page or execution note.

For optional payment automation, n8n documents a Stripe Trigger and signature verification in supported versions. Verify the event's origin and match the actual payment record before updating a project. That path is separate from writing an invoice reminder. The existing invoice follow-up article covers the latter.

Start with the chain whose output you can inspect in a minute: a draft, a photograph or a PDF. Add another transfer only when the first one reliably lands in the right place.

Fact-check notes and sources

Reviewed October 16, 2026. These are documented capabilities, not results from an account integration test. No prices or promised savings are used.

Starting list: a16z seventh edition. The hub handles its rankings and measurement limits; the workflows here are my proposed applications.
Transfers: n8n documentation for Notion, Gmail drafts, HTTP requests and webhooks.
Gamma: connectors and official n8n node, API access, generation request, status and export links and manual exports.
Specialist APIs: ElevenLabs speech conversion, Fireflies transcript query and Photoroom upload and editing parameters.
Models and assistant access: n8n Ollama model node, n8n OpenRouter model node, OpenRouter BYOK and Notion MCP.
Payment handoffs: Stripe invoice and payment link comparison and n8n Stripe Trigger authentication.

This post is informational, not legal, tax or medical advice. Product names identify the tools discussed. No affiliation is implied.

How Top AI Apps Fit Together for Solo Owners and Small Businesses