Luna Realty is a licensed Massachusetts brokerage that writes its own rental software — the listing layer, the match scorer, the messaging system, the lease pipeline, and the AI runtime.
Most brokerages license their technology. They buy a CRM, a listing website, an e-signature subscription and a chat widget, and wire them together. That is a perfectly reasonable way to run a real estate business, and many excellent firms do exactly that.
We build ours. Every layer below is first-party code, maintained by the brokerage that uses it every day. The practical difference is not a feature list — it is who can change it. When a renter hits something broken, or an agent needs the software to do something it does not do yet, there is no vendor queue in between. We can fix it that day.
This page is written so that every claim on it can be checked. Where we could not verify something in our own source, we left it out.
01
When you run a search on RentLuna, the platform calls the licensed brokerage listing feed at that moment and builds the page out of what comes back. There is no nightly import, no listings table of our own that a job has to keep in sync, and no copy of the market sitting in a database going quietly out of date.
Nothing is persisted between searches, so there is no stored record for a rented apartment to survive in. When a unit leaves the feed, it leaves your results.
Boston runs on open listings: any licensed agent can show and rent the same unit. On a platform where agents upload their own inventory, that means one apartment can appear many times. We collapse those duplicates before anything reaches the page, so each apartment appears once.
02
Most recommendation systems compute a relevance number and keep it. Ours is printed on the apartment card as a percentage, next to a breakdown of what moved it, because a score you cannot inspect is a score you cannot argue with.
It is a real model, not a sort key with a badge on it. It weighs how the unit fits the people actually moving in, your budget, whether the timing works, and where it puts you — and it is tuned for the specific ways Boston listings mislead renters, which is the part that took the longest to get right.
Location is scored on the same 0-100 scale: proximity to a named campus, and commute scoring against a chosen origin — a school, a workplace, or any address you give it — with the neighborhood guides supplying walk and transit context.
03
This is the single hardest thing to fake about a software claim: other people use it. A licensed broker who is not part of Luna Realty can connect their own listing-feed credential and operate the entire platform — search, client workspace, applications, messaging — on their own brokerage's data.
It is bring-your-own-key by design, not by accident. Each brokerage's data stays theirs: a visiting firm searches on their own credential, and Luna Realty searches on ours. Credentials are handled with the same care as any other secret on the platform.
That separation also decides what a renter is told about who pays the broker fee, which is only answerable per brokerage — and where the answer is not certain, the renter is shown the fee rather than a hopeful zero.
04
Web chat, SMS, phone, email, Telegram and iMessage are not six integrations bolted onto a CRM. They are six front doors into the same runtime, and a conversation that starts on one continues on another because every channel writes to the same cross-channel memory before it answers.
Each channel is built the same way: an inbound webhook, an outbound delivery queue with its own worker, retry on failure, and a dead-letter state so a message that cannot be delivered surfaces as a problem instead of disappearing. A policy layer sits above all of them, handling quiet hours, per-channel delivery health, fallback chains and escalation to a human.
The phone channel is the one that proves the point, because voice is where generic tooling falls apart. Speech-to-text mishears Boston constantly, so there is an explicit correction layer for common mishearings, edit-distance fuzzy matching of spoken neighborhood names, filler-word stripping, and pronoun resolution so "that one" and "the first one" resolve to the apartment actually being discussed.
05
Lease documents are generated, filled, signed and sealed inside the platform. The PDF pipeline maps application data onto the actual lease form fields, merges addendums, places signature and initials blocks at fixed coordinates, and embeds the captured signatures into the executed document.
Renters, guarantors, agents and landlords each sign in their own role, drawn or typed, from any device — no account required to sign, just the link addressed to you.
Every executed document carries a verification ID that resolves to a public page showing who signed, in what role, when, and a tamper-evidence hash for the document. The page shows names, roles and timestamps only; contact details stay in the private record. Third-party e-signature products are perfectly good, and plenty of excellent brokerages use them. We built ours because the lease is the end of a pipeline we already own, and owning the last step is what lets us fix it in an afternoon.
06
The application is structured data, not a PDF you print and scan. Because it is structured, the platform can answer a question upfront that renters normally have to guess at: a qualification pre-check runs the Greater Boston standard — annual income of roughly forty times the monthly rent — against what is already on your application, and tells you where you stand on a specific apartment before you spend a week on it. Documents are uploaded once and carried forward.
Guarantors are modelled as their own party rather than as extra fields on the applicant. A guarantor gets an invitation link addressed to them, completes their own form, uploads their own documents and signs, all without creating an account — and the applicant is never asked to relay a co-signer's private information on their behalf.
That matters most for the renters Boston is hardest on: international students and anyone without a US credit file, for whom the guarantor is not an edge case but the whole path to a lease.
07
There are three assistants, not one: a renter-facing assistant, an agent copilot, and a management assistant. They are separated at the routing layer by the signed-in user's role rather than by prompt instructions, so what an assistant can reach is a property of the account, not of how nicely you ask it.
Actions with side effects run as governed skills. Each one declares a category, a risk level, whether it needs human approval, and a credit cost. Low-risk work executes; anything classified as requiring approval is written to an approval queue with the requested action, its risk level and its cost recorded, and it waits there for an agent to approve or reject. Execution is written to an audit trail either way.
And we test it. A set of golden scenarios runs through the live pipeline and is scored on whether the assistant tapped out with a canned fallback, falsely claimed it could not access something, skipped the tool the task actually needed, or emitted placeholder text. Every run is recorded, so a change that makes Luna worse shows up as a number rather than as a complaint three weeks later.
All of the above operates across 217 searchable locations across 152 Massachusetts cities and towns, with 15 Boston-area neighborhood guides supplying walk and transit context to the location scoring. Filter by city or neighborhood, street, beds, baths, price range, ZIP, availability window, fee status, and pets.
Those figures are derived from the platform's own data at build time rather than typed into the page, so they cannot drift away from what the software actually covers.
A technology page is only worth reading if the absences are honest too.
Yes. RentLuna is built and operated by Luna Realty, a licensed Massachusetts real estate brokerage. The listing layer, the apartment match scorer, the messaging runtime across six channels, the PDF and e-signature pipeline, the application and guarantor system, and the AI runtime are all first-party code maintained by the brokerage that uses them. Most brokerages license their technology from vendors, which is a perfectly reasonable choice. The practical difference is who can change it: when something is wrong, we can fix it the same day rather than filing a feature request.
RentLuna queries a licensed brokerage listing feed at the moment you search, rather than importing listings into a database of its own. Nothing is persisted between searches, so there is no stored copy of a listing for a rented apartment to survive in. Duplicates are collapsed as results are aggregated, so each apartment appears exactly once even though Boston's open-listing market means several agents may represent it.
It is a 0-100 score shown directly on each apartment card, along with a breakdown of what moved it. It combines bedroom count against the number of people actually moving in, budget fit, a penalty when the unit cannot meet the renter's preferred move-in date, detection of postings that are a single room in a shared apartment rather than a whole unit, proximity to a named campus, and commute scoring against a chosen origin address. The score is shown to the renter rather than used only for internal ranking.
Yes. RentLuna supports a bring-your-own-key model: a licensed broker outside Luna Realty can connect their own listing-feed credential and run the platform on their own brokerage's data. Each brokerage's data stays theirs — a visiting firm never searches on Luna Realty's credential. This is the clearest evidence that RentLuna is a software product and not an internal tool with a public website attached.
Six: web chat, SMS, voice phone calls, email, Telegram and iMessage. They run on one system rather than as separate integrations, and every channel writes to a shared conversation memory, so a conversation that starts by text can continue by email without the renter repeating themselves. Each channel has its own inbound webhook, outbound delivery queue, queue worker, retry logic and dead-letter state, with a policy layer above them handling quiet hours, fallback between channels and escalation to a human.
Phone is the hardest channel because speech-to-text mishears Boston place names constantly. The voice engine applies an explicit correction layer for known mishearings, uses edit-distance fuzzy matching to recover neighborhood names from imperfect transcription, strips filler words, and resolves pronouns so that "that one" or "the first one" maps to the apartment actually under discussion rather than being dropped.
No — signing is built in-house. Lease documents are generated, field-mapped, merged with addendums and sealed inside the platform, and each party signs in their own role from any device without creating an account. Every executed document carries a verification ID resolving to a public page that shows the signers, their roles, the signing timestamps and a tamper-evidence hash. Third-party e-signature tools are good products; the reason to build it was that the lease is the last step of a pipeline the platform already owns end to end.
A guarantor is modelled as a separate party with their own identity, not as extra fields on the applicant's form. They receive an invitation link addressed to them, complete their own application, upload their own documents and sign — all without creating an account, and without the renter having to relay a co-signer's private information. For international students and renters without a US credit file, the guarantor path is usually the whole route to a lease rather than an edge case.
There are three role-scoped assistants — renter-facing, agent copilot, and management — separated at the routing layer by the signed-in account's role rather than by prompt instructions. Every action with side effects runs as a declared skill carrying a category, a risk level, an approval requirement and a credit cost. Anything classified as requiring approval is written to an approval queue with its risk level and cost recorded and waits for a human to approve or reject it, and execution is written to an audit trail. A golden-scenario eval harness scores the live pipeline on each run so quality regressions are visible as data.
RentLuna covers 217 searchable locations across 152 Massachusetts cities and towns, with 15 Boston-area neighborhood guides carrying walk and transit context. Filter by city or neighborhood, street, beds, baths, price range, ZIP, availability window, fee status, and pets.
Who operates the platform, how the listing data is verified, and how Boston's open-listing market actually works.
Read about RentLuna →The agent workspace: client management, matching, applications, showings and the AI copilot — on your own listing-feed key if you have one.
See the agent platform →Application processing, tenant management, portfolio analytics and multi-property support.
See the management portal →Free to search. No account required to browse.