Vendo Review (2026): Pricing, Features & Honest Verdict

Vendo is an open-source embedded agent layer that lets your own customers build features and micro-apps inside your SaaS product. The pricing page is the most honest we have read all year, and the category it sells into is still unproven. Best for: B2B SaaS teams buried in one-off feature requests. Price: From $49/mo (free plan: yes). Rating: 8.5/10.

What is Vendo?

Vendo answers a problem every B2B product team knows by heart. A customer asks for a view, a report, a small automation, something that would take a day to build and would serve exactly one account. It goes on the roadmap. It never ships. Multiply that by four hundred accounts and you have the backlog most SaaS companies actually run.

Vendo’s answer is to hand that work to the customer. You install an agent into your product, and your users describe what they want in plain language. Vendo assembles it live: a working micro-app with its own UI, state, scheduled automations, and the ability to take real actions through your API. Not a generated screenshot, not a chat answer, an app the customer keeps.

The setup is one command. Run npx vendo init and it reads your repository: your components and colours land in a theme file, your API endpoints become callable tools tagged read, write, or destructive, and your permission rules become a policy file where reads run free and destructive calls ask first. Anything Vendo builds wears your brand and runs sandboxed as the signed-in customer, inside permissions that customer already has.

The company behind it is Aisle Technologies, Inc., backed by Y Combinator and Founders, Inc. The code is on GitHub at runvendo/vendo under Apache 2.0, sitting at 454 stars since the repository opened on 30 June 2026. This was Vendo’s second Product Hunt launch, landing at #4 on 17 August with 336 points.

What are Vendo’s key features?

Micro-apps, not generated answers

The distinction Vendo draws is the whole pitch. Their own example: a customer of a finance app asks “what did I spend when I should have been asleep?” and gets back a dashboard with an hourly breakdown, the individual transactions, a saved alert at $100, and a rule that blocks DoorDash after midnight. That app runs nightly from then on. It has state, memory, and permissions. Most in-product AI gives you a paragraph and a chart that disappears when you close the panel.

One command that reads your codebase

vendo init detects your framework and scans routes, components, and theme. The output is three files you can read and edit: theme.json for colours, radii, fonts, density and motion; tools.json for your API surface with each endpoint classified by risk; policy.json for what runs unattended and what needs a confirmation. Making the extracted config plain JSON rather than a hosted dashboard is the right call, because it means the thing governing what an agent can do in your product lives in your repo and goes through your code review.

Nightly verification against 16 real open-source apps

Vendo runs init every night against a set of real production codebases, and publishes the results on the homepage. The named ones include cal.com, dub, formbricks, papermark, twenty, rallly, openstatus, umami, skateshop, inbox-zero, teable, and invoify. Publishing a live pass or fail for your onboarding step against apps you do not control is a claim you cannot quietly walk back, and we have not seen another developer tool do it. It is also the honest way to say “this works on real code”, because you can go clone any one of those repos and try it yourself.

The same app runs in Claude, ChatGPT, and on a schedule

An app a customer builds is reachable from three places: inside your product as an overlay, sidebar, panel or remixed into your own components; from Claude or ChatGPT, still wearing your brand; and on a cron schedule with the permissions the user granted. The scheduling piece is the one that turns this from a novelty into infrastructure. “Every Monday at 7, chase the clients with missing documents” becomes software that runs itself, as that user, and reports back.

How much does Vendo cost?

Vendo starts at $49 per month, and there is a free plan that does not expire. Four tiers are published. Free is $0 forever and includes $5 of usage a month, roughly 100 agent turns, with a hard stop at $5 that produces an error rather than a bill. Pro is $49 a month and includes $49 of usage, adding sharing, publishing, live collaboration, and 90-day history. Teams is $499 a month and includes $499 of usage, adding approvals, session replay, org activity and insights, with unlimited retention. Enterprise is custom: committed usage, SAML SSO and SCIM, an SLA with credits, and deployment on your own infrastructure. Annual billing takes Pro to $490 a year and Teams to $4,990, about two months free.

Underneath the plans there are exactly four published rates, identical on every tier including Free. Model tokens run at $1.15 per million input and $5.75 per million output on the cheap models, or $2.30 and $11.50 on the default. Sandbox time is $0.01 per minute of wall clock. Storage is $0.25 per GB-month. Automation runs are $3 per thousand. Vendo puts real numbers next to those rates: a typical agent turn is about 5 cents, an app build is $1 to $2.

Compare that to how this category usually prices. CopilotKit charges $39 a month for Pro and $100 per seat on Team. Retool prices per internal user, at the time of writing €9 for Team and €46 for Business on annual billing. Paragon, the closest embedded-integration comparison, publishes no prices at all. Vendo is the only one of the four that will tell you what a single operation costs before you sign up.

Is Vendo’s pricing model actually a good deal?

Read the plans again and notice something unusual. Pro is $49 and includes $49 of usage. Teams is $499 and includes $499 of usage. The price of the plan is the usage allowance, exactly, with no markup and no rounding.

That is refreshingly clear, and it also means the subscription is a minimum spend rather than a discount. If you use $12 of Vendo in a month on Pro, you paid $49. If you use $80, you pay $80, because the bill is whichever is higher. You are not buying cheaper tokens by upgrading. You are buying features: publishing, collaboration, retention, approvals, and the right to run past your included dollars at all.

Whether that is good depends entirely on your volume. A team that is live in production and burning $200 a month gets Pro’s features for free, since usage covers the plan price. A team evaluating with a few dozen turns a week is paying $49 for publishing rights. Vendo’s own free tier is generous enough that you can measure this before committing, and the spend cap is on by default at $100 on Pro and $1,000 on Teams, with warnings at 50, 90 and 100 percent. Turn overage off and usage simply pauses at what your plan includes.

One gap worth naming. The models are called vendo, vendo-apps, vendo-extract, vendo-review and vendo-judge, and nowhere does Vendo say what they actually are. The rates are published so your cost is predictable, but you cannot tell which model is writing code against your API, and you cannot compare the rate to the underlying provider’s list price. Every rate has a bring-your-own-key path that reads $0, which is the pressure valve, but it only helps if you already know what you want to run.

How open source is Vendo really?

Properly, as it turns out, and that is worth checking rather than assuming. The runvendo/vendo repository carries a plain Apache 2.0 licence with no additional terms bolted on, which is the permissive end of the spectrum and puts no restriction on running it commercially or offering it as a service. The repo was pushed to the morning we wrote this, so it is not a marketing mirror of a closed product.

There is one seam. The software is Apache 2.0, so nothing stops you self-hosting it, and the pricing FAQ explicitly tells you to self-host the open-source store on your own Postgres to avoid the storage rate. But “deploy on your infrastructure” appears in the feature table as an Enterprise-only row. Both things can be true, since the licence permits self-hosting while support and the supported deployment path are sold, but a reader skimming the table will conclude self-hosting is gated when the licence says otherwise. Clarifying that row would cost Vendo nothing.

Who is Vendo best for?

Use Vendo if you run a B2B SaaS product with a real API, a customer base that keeps asking for small custom things, and an engineering team that will never get to those requests. That is the shape the product is built around, and the maths works: one Pro subscription against a backlog of forty one-account feature requests is not a close call.

Use it if you want in-product AI that produces durable software rather than chat answers, and if you care that the config governing agent permissions lives in your repository rather than someone’s dashboard.

Skip Vendo if your product has no API worth exposing, because there is nothing for generated apps to act on. Skip it if you are consumer-facing with a simple surface, where per-customer customization is a solution to a problem you do not have. And skip it for now if you need a track record: the repository is seven weeks old, the category has no incumbent, and you would be betting on an idea as much as on a product.

Best Vendo alternatives

CopilotKit is the closest open-source comparison, an in-app AI copilot framework you embed in your own product. Its Developer tier is free forever for one developer, Pro is $39 a month, and Team is $100 per seat per month up to five seats. CopilotKit gives you the building blocks and expects you to define the experience. Vendo ships the experience and reads your codebase to configure itself. Pick CopilotKit if you want control over every interaction.

Retool attacks the same backlog from the other end. Instead of letting customers build, your team builds internal tools fast and exposes what it must. Free covers up to five users with 500 workflow runs a month, and paid tiers ran at €9 and €46 per internal user on annual billing when we checked. Retool is mature, proven, and priced per human. It does not solve per-customer customization, only per-team velocity.

Paragon is the embedded iPaaS answer: you ship integrations to your customers without building each one. It covers connecting your users’ tools, which is one of Vendo’s four pillars, and does not attempt the app-building half. Paragon publishes no pricing, only a free trial and a demo booking, so budgeting means a sales call. That contrast is the sharpest argument for Vendo we can make.

Final verdict: is Vendo worth it?

Vendo is the best-argued product we have reviewed this month, and the pricing page is a small masterclass. Four rates, published identically on every plan including free, each with a bring-your-own path to $0. A free tier that errors instead of billing. Spend caps on by default. A written commitment that the rates are never redefined and that existing subscribers keep their terms with 30 days’ notice. Nightly public verification of the onboarding step against apps the company does not control. Almost nobody does this, and the ones who do usually have something to hide elsewhere. We looked and did not find it.

The risk is not the company, it is the category. Nobody has proven that end customers of a B2B product will build their own software, or that the accounts asking for custom features will accept building it themselves instead of being served. If that behaviour lands, Vendo is early and well-built. If it does not, you have a very elegant answer to a question your customers were not asking. The good news is that finding out is nearly free: the $5 monthly allowance is real, the CLI runs against your actual repo, and the config it writes is three JSON files you can read in five minutes. Run vendo init on a branch before you decide anything. It is the cheapest way to learn whether your API is ready for this at all.