AI ASSISTANT FOR PROPERTY SOFTWARE
Build an assistant for your property management software. Ask in plain words for cited answers or an API request preview. Review the operation and its details. Previews do not run operations.
What Orrin does
For product and engineering teams at property management software companies who want an assistant in their product without building an AI team.
Your users ask in plain words and get an answer from your documentation, with a citation they can open.
How do I waive a late fee for a tenant?
Create a work order, schedule an inspection, send a rent reminder: each request is matched to one of your API operations and shown as a preview, with the values it read and its confirmation level. This release does not execute the operation.
Create a work order for the leaking tap in 4B.
Every request carries the signed in user’s role and scopes in a short lived token signed with your own key, so the assistant works within that user’s permissions and no one else’s.
Signed in as a leasing agent: units and tenants, read only.
Three confirmation levels
Previews a lookup request without calling your API. Previews a change to one record and describes the confirmation it would require. Previews a sensitive operation and describes the typed confirmation it would require.
Sensitive
Sensitive request, prepared for review
“Remind everyone with a balance”
Who it is for
Orrin is for companies that build property management software, and each of their teams gets something different from it.
Put an assistant in your product without building an AI team. You decide what it answers from and which actions it may prepare.
Let your documentation answer the questions it already covers, and see the ones it does not, listed in your users’ own words.
Import the OpenAPI file you already have, sign a short lived token for each user with your own key, and review each proposed operation in the console.
How our AI works
Orrin turns a vendor’s documentation and API into cited answers and previews of the actions a user asks for. Every part below runs in the product today.
Guides, pasted text, files and public pages are split into passages, a new one at every heading and long sections in overlapping pieces. Each passage keeps its heading and is indexed for its own workspace only, so an answer can point to the exact section it came from.
Orrin runs all-MiniLM-L6-v2, a small open source language model, inside its own server. It turns every passage, question and API operation into a list of numbers that point the same way when two texts mean the same thing, even when they share no words. No text is sent to an outside AI service for this.
Every question is ranked three ways at once: by meaning with that model, by BM25 keyword scoring for exact terms, and by hashed word and phrase vectors for overall similarity. The three are blended into one score, with meaning counting most.
A passage counts as evidence only when its meaning is close enough to the question’s. An answer quotes at most two sentences whose own meaning is close, each with a numbered citation that opens the source passage. When nothing qualifies, Orrin says the documentation does not cover it rather than guessing.
A plain request is compared by meaning with every API operation you switched on, and the closest is previewed with its confirmation level. When two operations are equally close, Orrin asks which one you mean instead of choosing.
The preview proposes the values it can read from the request, such as a unit, a date, an amount or a reason, only for parameters your operation defines, and lists the required ones still missing. This release does not execute the operation.
Each imported operation gets a suggested level. Reads are Read; anything that deletes, moves money, sends messages or changes records in bulk is Sensitive; everything else is Write. You review every label before an action is exposed.
Questions the documentation cannot answer are counted on the Usage page and can be raised as support tickets, so your documentation team sees exactly what is missing.
The engine is scored on a fixed set of questions and requests it was never tuned on, written against two made up companies. The results, failures included, are on the evaluation page.
Good to know. Answers come from the documentation Orrin is given and can still be incomplete or wrong, so check the cited source before acting on one. Nothing the assistant says is legal, financial, tax or other professional advice.
Case study
Brickwren is an example property management platform, not an Orrin customer: 14 help articles and an API with 14 operations. Both went through Orrin’s engine, the same code a workspace runs. This is what came out, word for word.
The inputhelp.brickwren.example
Help center
API · OpenAPI 3
5 Read3 Write6 Sensitive
Orrin saw
14 articles, indexed as 50 passages. 14 operations, labelled 5 Read, 3 Write and 6 Sensitive from their methods and wording. Brickwren switched on 10 and kept 4 off: record a payment, cancel a work order, send a renewal offer and refund a security deposit.
The outputWord for word
Ask
How do I waive a late fee for a tenant?
To waive a late fee, open the tenant's ledger, select the fee and choose Waive fee.1 You need the Waive fees permission to do this.1
Late fees
To waive a late fee, open the tenant's ledger, select the fee and choose Waive fee.
Late fees
The late fee is added to the tenant's ledger the morning after the grace period ends.
Late fees
Late fees
Late fees
Late fees are set for each property in Property settings, under Late fees.
Act
Waive the late fee for the tenant in 4B
waiveChargeSensitivePOST /charges/{charge_id}/waive · Waive a charge
No values found in the request. Still needed: charge_id, reason.
Get started
Connect your documentation, import your API and scope a test identity, then try cited answers and action previews in the console.
Add a public help page, paste text or upload Markdown, HTML, text, JSON and CSV.
Orrin indexes the content in your workspace so answers can cite the source passages.
Import an OpenAPI specification.
Review candidate actions, choose which ones are exposed and label them Read, Write or Sensitive before trying an action preview.
Generate a signing key, approve the sites your product runs on and mint a ten minute test token with a role and scopes.
Then have your backend sign one for each signed in user.
Ask real questions, open each citation and check every action preview in the console.
Then add one script tag, and your users get the assistant in your product.
Actions
The requests property teams make every day, each with its confirmation level. Import your OpenAPI file and Orrin previews requests like these against your own operations.
Matches the question to your tenant lookup operation and shows the call it would make, with the unit read from the request.
Matches the request to your units operation. Only the filters that operation defines can be proposed, each read from the request.
Matches the request to your renewal operation, proposes the unit from the request and lists every required value still missing.
Matches the request to your work order operation, proposes the unit and description from its words and lists any required values still missing.
Matches the request to your assignment operation and proposes the work order number from the request. A vendor it cannot read is listed as missing.
Matches the request to your inspection operation and proposes the unit, the type of inspection and the day from its words.
Matches the request to your statement operation and shows the call it would make, listing any required value it could not read.
Matches the request to your reminder operation. The preview describes its Sensitive confirmation level and lists any required values still missing.
Matches the request to your notes operation and proposes the note from its words for the user to review.
Safety
Every action request becomes a preview: the operation Orrin matched, its confirmation level, the values it read from the request and any still missing. This release does not execute the operation. A preview never calls your API.
Orrin suggests a level for every operation you import, and you review it. Nothing is exposed until you switch it on, and only exposed operations can be matched.
Your backend signs a token for each user that lives for minutes and carries their id, company, role and scopes. The console generates your signing keys and verifies the tokens they sign, so every request belongs to one user and never to a shared service account.
Documents added and removed, questions asked, previews shown, keys created and revoked, members invited: each is recorded with who did it and when, and the whole log downloads as CSV.
The acceptable use policy forbids using the assistant to filter, rank, describe or steer units, applicants or tenants by any protected characteristic, and forbids mapping actions for screening decisions, evictions or legal advice.
Your documents and questions are never used to train models. Deleting a document removes its passages from the index at once.
The Security page lists every practice and whether it is in place.
Analytics
The console counts questions, grounded answers, unanswered questions and action previews every day, and exports them as CSV. Unanswered questions show the gaps in your documentation.
Every question asked in the workspace, counted by day.
Answers backed by a cited passage from your documentation.
Questions your documentation could not answer. This is your documentation backlog.
Requests matched to an operation you exposed, counted by day.
Any unanswered question can become a ticket, with replies and a status.
Daily counts, conversations and the audit log, downloadable for your own reporting.
Developers
Your backend signs a short lived token for each user with a key generated in the console, and the assistant checks it on every request, so each answer and preview belongs to one user and stays within their scopes.
// Your backend, whenever the assistant asks. The token lives ten minutes.
app.get("/api/orrin-token", requireSession, async (req, res) => {
const token = await new SignJWT({ company: req.user.companyId, role: req.user.role, scopes: req.user.scopes })
.setProtectedHeader({ alg: "EdDSA", kid: ORRIN_KEY_ID })
.setSubject(req.user.id)
.setIssuedAt()
.setExpirationTime("10m")
.setAudience("orrin")
.sign(privateKey);
res.type("text/plain").send(token);
});About the founder and team
I’m building Orrin, an AI assistant that lives inside property management software.
Property platforms hold a lot of answers and a lot of routine work. Orrin lets their users ask in plain words. It answers from the software vendor’s own documentation, with a citation they can open, and prepares tasks in the platform, showing exactly what will change so the user can confirm it first.
Pricing
Free
For building and testing in the console.
200 test conversations a month
$299 a month
Billed monthly
For teams starting with a focused set of actions.
3,000 conversations a month
$999 a month
Billed monthly
For growing platforms with more operations to preview.
15,000 conversations a month
$2,499 a month
Billed monthly
For larger user bases.
50,000 conversations a month
Custom
Custom volume and terms for the largest platforms.
Custom volume
Get started