Tektolo Privacy Policy
Effective: November 14, 2026 · Site: https://tektolo.com
Changed on November 14, 2026: our pages now send a reading signal, described in the table below.
Contents (17 sections)
Corey Kozak, an individual in North Carolina, USA, operates Tektolo ("we"). This policy explains what the website, web tool, REST API and MCP endpoint (the "Service") collect. The Service has no accounts or sign-in. It is free to use; an optional paid lane for automated clients accepts crypto payments (see Paid calls below).
Short version: we do not log what you send us, we do not use cookies, advertising trackers or third-party analytics (our own pages send our server one signal that a person is reading them; see the table below), and we keep one anonymized record per estimate to run and improve the Service. Agencies in our pilot are the exception: they get a cookie that tells us which agency they are, and what they do is kept under their agency's name (see Pilot agencies). On the property turnover tool we also keep the text you type (your notes, or the list of issues an assistant sends), with contact details and access codes removed, together with the findings it became, for up to 5 years, to improve our matching and measure its accuracy. You can opt out (see Notes and matching). We do not sell data and do not use it to train third-party AI models. On the property turnover tool, whether you use the web page, the API or an AI assistant, your notes (or the issues an assistant sends) are sent to Microsoft Azure OpenAI to read, every line, exact matches to our list of standard repair items included (see AI reader). If you upload an inspection report or photos to the property turnover tool, the pages are converted to images and sent to Microsoft Azure OpenAI to read; we do not keep the file, the images or the text read from them (see Document upload).
What we collect, what we never collect, how long
| What | How long | |
|---|---|---|
| Collected: the estimate request itself | Held in memory to compute your result. Not logged, not stored. | Discarded when the response is sent |
| Collected: usage event (log line per call) | Channel (web, REST or MCP), capability and operation, status, duration, hashes of the request and result, software and rules versions, a short error message if it failed, and an anonymous daily fingerprint (see below) | Up to 30 days in Azure Log Analytics |
| Collected: page view (log line per page) | The page address without its query string (none for a page that does not exist), its status, and the same where-from details as usage events (below) | Up to 30 days in Azure Log Analytics |
| Collected: reading signal (at most one log line per page load) | A small script of ours on each page tells our server, once, when the page has been open and visible for a few seconds and you have done something on it, such as using a mouse wheel, moving or pressing the pointer, touching the screen or pressing a key. It says only that this happened: not which key, not what you typed, clicked or pointed at, and not how far you scrolled. The log line holds the page address without its query string, the campaign tag from the link you followed if there was one, the category of software that made the request, and the anonymous daily fingerprint (see below). The script sets no cookie and keeps nothing in your browser, and without JavaScript nothing is sent. We use the signal to tell visits by people from automated fetches, such as email security scanners that open the links in an email, so that our counts of visits, including visits from links in our own emails, count people. | Up to 30 days in Azure Log Analytics |
| Collected: where a visit or call came from (on page views and usage events) | A campaign tag from the link you followed (ref, utm_source or utm_campaign, only a short plain word), the name of the site that linked to us (the host only, never the full address), a category for the software that made the request (for example "browser", "curl/http-client" or a named search or AI crawler such as "Googlebot" or "ClaudeBot"; never the full user-agent string), and for AI assistants the client name and version they announce over MCP (for example "claude-ai/0.1.0") |
Up to 30 days in Azure Log Analytics |
| Collected: anonymized estimate record (one per estimate) | Channel; ZIP3 (first 3 digits of the ZIP); unit type, bedrooms, bathrooms, size rounded to the nearest 100 sq ft, build year rounded to the decade, whether there is carpet; standards chosen (profile, price tier, which options were set); finding codes, areas, extents, whether a quantity was given, and how each finding was matched; counts of lines and trades; expected totals; referral rule ids; metro area; whether a rate card was supplied (yes/no) and number of approve/decline decisions; hashes; the daily fingerprint; time rounded to the hour | Up to 730 days, then automatically deleted |
| Collected: text of findings we could not match | If a finding's description matches nothing in our list, or a finding's code is not one of our codes (free-form text), a scrubbed copy is kept: likely names removed, lowercased, with links, email addresses and all digits removed, cut to 120 characters (80 for a free-form code). This is the only free text we keep. Scrubbing is automatic and its name removal is best-effort, so please do not type personal details into findings. | Same record, up to 730 days |
| Collected: the text you type into the property turnover tool (unless you opt out) | The whole text of your notes, or the list of issues an AI assistant or other software sends; and for each finding, the words it came from, what our matcher or the AI reader suggested, and what you finally estimated with (the issue, area, extent and quantity, and whether you kept, changed, removed or added it). Before anything is stored we remove email addresses, phone numbers, street addresses, apartment and unit numbers, gate, lockbox, door, alarm and Wi-Fi codes and passwords, long numbers, links and likely names. This is automatic and best-effort, so please do not type tenant names, contact details or access codes. Also kept: software and rules versions, the daily fingerprint, a network tag (see below), the time to the hour, and the id and hash of the estimate the findings went into. On the web page, a finding that was already in an earlier estimate in the same visit also keeps the id of the first estimate it was in, whether you changed it, removed it or kept it. These ids can link the record to those estimates' anonymized estimate records. See Notes and matching. | Up to 5 years, then automatically deleted |
| Collected: feedback you choose to send | What you type into a feedback form (see Feedback below): the message, optional source, reporter and agent labels, the page it was sent from, a rating and reasons if you give them, an email address only if you type one, and, when sent from an estimate result, a few identifiers of that estimate (hashes, software and rules versions, metro area, expected total; never your request or notes). Also the daily fingerprint. Now and then we also copy a public post about us from a third-party site into these records, without the author's handle or link (see Agent and relayed feedback below). | Up to 730 days, then automatically deleted |
| Collected: optional pay-question answers | What you choose on the "Would you pay for a tool like this?" question under a tool result: per move-out, monthly or not at all; a dollar amount if you type one; whether you would rather pay per result from a prepaid balance, pay on a monthly plan, or have no preference; and a minimum top-up amount if you type one. With it we keep the tool and the page it was shown on, the time to the minute, hashes of the request and result and the software and rules versions of that result (never the request itself), whether it came from the web page or the API, and the daily fingerprint (see below). No message and no email address are kept with it. We take no payment and store no payment details. | Up to 730 days, then automatically deleted |
| Collected: a pilot application (only if you apply) | What you enter on the pilot form: your name, work email, company, role, number of doors, state, move-outs a month, how you price turnovers today, your software and anything you write about move-outs (likely names and contact details in that text are removed), that you agreed to be emailed about the pilot, the campaign tag of the link you followed, the time and the daily fingerprint | Up to 3 years, then automatically deleted; sooner if you ask |
| Collected: an unsubscribe request (only if you unsubscribe) | A one-way hash (SHA-256) of the email address you give on the unsubscribe page or in a "stop" reply to our email, how it reached us and the time. We do not store the address itself; the hash cannot be turned back into it, but anyone who already has the address can compute the same hash and check it against the list, which is how our own check works. See Unsubscribing | For as long as we send outreach emails; not deleted on a timer, because it is what stops us emailing you. Deleted sooner if you ask |
| Collected: pilot agency records (only if your agency joins the pilot) | What your agency does with the Service during the pilot, tagged with your agency: each estimate and settlement you run, the pages you view and the tools you use, the findings our tools read or suggested and what you kept or changed, your ratings, what each job actually cost and any invoices you choose to upload, your answers to the optional weekly questions and the baseline question, and the exit interview. See Pilot agencies | Up to 3 years after each pilot record was last written, then automatically deleted (the correction log: up to 5 years; estimate records and ratings: up to 730 days; log lines: 30 days); sooner if you ask |
| Collected: paid-call records (x402) | Only if you use the paid lane: the wallet address that paid, the transaction id and network, the amount, which operation was called, a hash of the result, the campaign tag from the link you used if any (ref), and the time to the minute. See Paid calls below. |
Up to 730 days, then automatically deleted |
| Collected: a deposit settlement request | State, city, amounts, dates, deduction categories and documentation flags, held in memory to compute the result. Not stored; no per-settlement record is kept. Deduction descriptions are never logged, hashed or stored; they are printed only on the letter you download. Only the standard usage log line above is kept (capability, status, duration, hashes, daily-rotating fingerprint). For an agency in our pilot, we keep a record of each settlement it runs (see Pilot agencies). | Discarded when the response is sent (a pilot agency's settlement record: see Pilot agencies) |
| Collected: a security deposit check request | State, city, amounts, dates, what the landlord returned and when, kinds of charges and wear answers, held in memory to compute the result. Not stored; no per-check record is kept. Only the standard usage log line above is kept. | Discarded when the response is sent |
| Collected: a summary pasted into the verify page | The text, held in memory to compare it with the code. Not stored. | Discarded when the response is sent. |
| Backup copies of deleted records | When a stored record is deleted, by us or automatically at the end of its time above, Azure Storage keeps a recoverable copy of it (soft delete and earlier versions). We use these copies only to undo an accidental or malicious delete. | Up to 30 days after the record is deleted, then removed automatically |
| Collected: a masked test copy | Now and then we copy a selection of the stored records described above into a separate test environment, to reproduce a problem and to build, test and evaluate the Service, including our own matching. Before copying we drop the daily fingerprint, the network tag, IP addresses, user agents, email addresses and payer wallet addresses, and we mask contact details, addresses, unit numbers, access codes and likely names in the remaining text. The copy is readable only by the test environment's app and the copying job. | Up to 30 days, then automatically deleted |
| Standard hosting data | Like any web service, our host handles connection details such as IP address and request path to deliver the page and keep the Service secure. Our own code does not store or log raw IP addresses and our hosting platform logs may contain them (see note below). | Up to 30 days |
| Never collected or stored | The text of requests that opted out, location labels, client references, full ZIP code, street address, IP address or user agent in our records, names, phone numbers or other contact details (we remove them from the text we keep; the exceptions are an email address you choose to type into a feedback form, see Feedback, the details you give when you apply to the pilot, see Pilot applications, and the contact details on a pilot agency's record, see Pilot agencies), rate-card prices, vendor names or contacts, decision comments, the files, photos and page images you upload, and the text read from them (only the findings you keep become part of your estimate), except the invoices a pilot agency chooses to upload with its actual costs (see Pilot agencies), and any text typed into the deposit tools | n/a |
| Cookies and trackers | We set cookies only for agencies in our pilot. When someone at a pilot agency opens the agency's pilot link, the site sets a tk_pilot cookie. It holds the agency's pilot code (a short form of the company's name, the link's last day and a signature) and nothing about the person. While it is set, the pages you view and the tools you use from that browser are tagged with your agency (see Pilot agencies). It expires at the end of the link's last day, and the Leave button on your pilot page or your browser's site-data setting removes it sooner. If someone opens the agency's exit interview link, the site also sets a tk_interview cookie. It works only on the interview pages, holds the interview link's code and expires when that link does, 4 weeks after the pilot ends. Nobody outside the pilot gets a cookie from us. We use no third-party analytics, no advertising or social trackers, and no third-party fonts or scripts. Our own page-view, reading-signal and usage log lines above are the only measurement. They hold no IP address or full user agent; a pilot agency's lines carry its agency tag. If you pin a tool or open one, the site remembers it in your own browser's local storage (the tool's address and when you opened it) to show "Your tools"; it is never sent to us, and the Clear button or your browser's site-data setting removes it. |
Until you clear it (pilot cookies: until the pilot or interview link ends) |
Note on hosting logs: the platform in front of our code (Microsoft Azure Container Apps and its web server) may write ordinary access logs that include the connecting IP address. We do not use these to identify you. We use them only to keep the Service running and secure, and they are deleted within 30 days.
The anonymous fingerprint
To count usage and spot abuse, we compute a one-way hash of your IP address and browser or client identifier, mixed with a secret value that changes every day and every deployment. The result looks like fp_3a9c… and cannot be turned back into an IP address. It cannot be linked from one day to the next. It appears in the usage events, page views, reading signals, estimate records and feedback records above. Rate limits use a separate one-way hash of the IP address alone, which is held only in memory (for a few minutes for the per-minute limits, and until the end of the day, UTC, for the daily free-call and connector URL counts) and is never stored or logged.
The network tag
To count how many different people agree on how a phrase should be matched, without knowing who they are, the text records described in Notes and matching also carry a network tag: a one-way hash of the first three parts of your IP address (the block of 256 addresses it belongs to; for IPv6, the first 48 bits), made with a secret key we keep private. Everyone on that block gets the same tag, and it cannot be turned back into an address. Unlike the fingerprint it does not change from day to day, so that one person who comes back on several days still counts once. We use it only to weigh agreement and to spot abuse; it is never published and never used to identify or contact anyone. It is not stored when you opt out. If you use a connector key, a one-way hash of the key's id is used the same way.
Feedback
The site has feedback forms: a rating under each estimate result ("Did this estimate look right?"), a "Tell us what this was" form beside findings we could not match, a "Request an improvement" form on each tool page, a "Suggest a tool" form on the home page and a general feedback page. Using them is optional. Under a result, a folded question asks whether you would pay for a tool like this. Answering it is optional.
- What we store. The text you write (links are replaced by "[link]" and messages are cut to 2,000 characters), the page it was about, your rating and the reasons you ticked, and the time to the minute. If you use the unmatched-finding form we keep the finding text exactly as you typed it, because you are sending it to us on purpose. If you send a rating from an estimate result we also keep a few identifiers of that estimate: hashes of the request and result, software and rules versions, the metro area and the expected total. We do not keep the request itself, your notes or your ZIP code. We also keep the anonymous daily fingerprint (see below), used for spam control.
-
Agent and relayed feedback. People and agents may send feedback directly through our API. Optional source, reporter and agent-name labels describe the submission; they are self-reported, not verified identities. We remove links and mask likely personal details from these labels as we do from feedback text. An operator may also copy a public Moltbook feedback quote into our private feedback records. We mark that copy as third-party, relayed feedback and keep its original source time; we do not copy the author handle or source link. These records have the same purpose and access described in this section, and are deleted up to 730 days after we copy them. Labels are not kept with pay-question answers.
-
The pay question. Your answers are stored as described in the collection table above. Nothing is charged and no price is quoted. An answer does not buy anything, commit you to a price or start a subscription, and it is not an offer or a price from us: what we build and charge, if anything, may differ from any answer. We use the answers to judge demand and to decide what to build and what to charge for. If you open the tool through a pilot agency's link, the answer is tagged with your agency like any other feedback you send from a result. A second answer to the same result from the same browser on the same day is kept as a second record and counted once.
- Email is optional. We store an email address only if you type one into the form. We use it only to reply to you about that feedback, never for marketing, and we do not share it.
- Please do not include personal details about yourself, your landlord, tenants or other people. If you do, we may delete them.
- Who sees it. The records sit in a private Azure Storage container. A daily summary of new feedback is posted to our project issue tracker on GitHub with email addresses masked in every field (for example
a***@example.com). Pay-question answers are not copied into that summary, which only says how many came in. We read them from the private container with our administrator account. Our own computer also keeps a copy of each answer (the choice, the amounts, the result hash and the daily fingerprint) in a private counting file, which drops an answer when storage deletes the record, and in our private daily sweep reports. Those copies stay on our computer and are not shared. - How long. Up to 730 days, then automatically deleted. To have your feedback deleted sooner, email us the approximate date and what you wrote.
- Why. To fix problems, tune our rules and decide what to build next. Not used to train third-party AI models.
Paid calls (x402)
If you or your agent pay for a call through the paid lane, we store the paying wallet address and the transaction id (with the network, amount, operation, a hash of the result, the ref campaign tag if any, and the time to the minute) for accounting, tax records, abuse prevention and handling payment questions. We also use these records for aggregate demand statistics; anything we publish is counts and totals only, never wallet addresses or transaction ids.
- How long. Up to 730 days, then automatically deleted. We may keep records longer only where the law requires it for accounting or tax.
- Blockchains are public. The payment itself is recorded permanently on a public blockchain (Base), where anyone can see the paying and receiving addresses and the amount. That is outside our control and cannot be deleted. A wallet address can sometimes be linked to a person by others.
- Payment facilitator. Payments are verified and settled by a third-party x402 facilitator (on the test network, the public x402.org facilitator; on Base mainnet, Coinbase Developer Platform), which processes the payment under its own terms and privacy policy. We do not receive your name, email or card details, and we never hold your keys.
Notes and matching
The property turnover tool turns your words ("BR2 carpet trashed by dog") into standard repair items. Every time someone checks and corrects those items, we learn how people really describe a unit, so we keep that text. This applies only to the property turnover tool; the deposit tools keep no text.
- What we keep. The text you submit (the notes box on the tool page, or the list of issues an AI assistant or other software sends us), what each finding was matched to, and what you finally estimated with, including findings you changed, removed or added yourself. Before storing, we remove contact details and access codes as described in the table above. The removal is automatic and best-effort: please do not type tenant names, contact details or gate, lockbox or alarm codes.
- Why. To improve how our matcher reads your words, to measure how accurate it and the AI reader are (about one note in ten is set aside only for that measurement), and to spot abuse. A correction never changes results for anyone else by itself: phrasings that several independent people agree on are reviewed by us before they are added to our matching rules.
- What may become public. A short phrase that many people use, such as "carpet trashed by dog", may be added to our published list of matching phrases. Never a whole note, never anything that was masked, and never anything that identifies you.
- Opting out. On the web, tick "Don't keep my notes" before you add your notes to the list or estimate. Through the API or an AI assistant, send
"retain": falsewith the estimate. Then we keep none of your text: the anonymized estimate record keeps only codes and counts (not even the scrubbed text of unmatched findings), and the AI reader's answer is not stored. We count how often people opt out, as a number only. - How long. Up to 5 years, then automatically deleted.
- Who sees it. The records sit in a private Azure Storage container, readable by our administrator account. For up to 30 days a masked copy of some records can also sit in our test environment (see the table above). They are not sold, not shared, and not used to train third-party AI models.
- Deleting it. There are no accounts, so we cannot look up one person's notes. If you want something removed, email support@tektolo.com with the approximate date and some of the text you typed, and we will search for it and delete the matching records.
How we use data
To operate and secure the Service (rate limiting, abuse prevention, debugging); to understand demand and improve accuracy of our rules, matching and coverage; to build, test and evaluate the Service, and to train and evaluate our own matching and models (never third-party ones); and to produce aggregated statistics, such as published cost-by-area pages. We do not sell or share data for advertising, do not build profiles of people, and do not use data to train third-party AI models. We may disclose information if required by law or to protect the Service from abuse.
AI reader
The property turnover tool reads free text the same way wherever it comes from. On the web page, that is the notes box: nothing in it is sent anywhere until you press "Add to the list" or estimate. Through the REST API, the paid lane or an AI assistant (MCP), it is the plain-English issues of a quick estimate. Then:
- What is sent, and to whom. Your text is sent, with our list of standard repair items, to Microsoft Azure OpenAI (the gpt-6-luna model, run by Microsoft for us; Microsoft processes the text inside its US data zone, in Azure data centers in the United States). It returns suggested findings. On the web page you check and edit them; through the API or an assistant they come back in the estimate marked as read by AI, with their lines flagged for review. Prices are always worked out by our own rules, never by the AI model.
- What we keep. Unless you tick "Don't keep my notes", the text and the findings it became are kept as described in Notes and matching. So that the same text need not be sent to the model again, the AI reader's answer is also stored, found by a one-way hash of the text; an answer that repeats contact details is not stored, and neither is the answer for an opted-out request. Our usage log line for each reading holds counts only (how many parts were sent, how many findings came back, timing, the model and software versions) and the anonymous daily fingerprint, never the text.
- What Microsoft does with it. Microsoft processes the text under its Azure OpenAI terms as our processor. It is not used to train AI models, and it is not shared with OpenAI. If Microsoft's automated abuse monitoring flags it, Microsoft may store it in the United States so that authorized Microsoft employees can review it.
- Please do not type tenant names or other personal details into the notes box or the issues you send. Describe the unit, not the people.
- Findings you keep become part of your estimate, which is handled like any other estimate described above; the anonymized estimate record notes that a finding was read by the AI reader.
The AI reader takes up to 12,000 characters at a time (through the API, all the issues of one request together); longer notes are not sent to the AI model, and the tool asks you to split them. Notes it has no daily readings left for, and notes it cannot reach the AI model for, are read without the AI model: only words that exactly match an item on our list are matched (our safety checks still read them), and the rest is listed as not read and priced at nothing. Findings sent with a code from our list are never sent to the AI model. The pilot's exit interview is the only other place where text you type goes to an AI model (see Pilot agencies).
Document upload
The property turnover web tool can read findings from a move-out inspection report you upload: a PDF, a Word, Excel or PowerPoint file, a CSV file, or up to five photos of paper pages. Software can also send a document to our API with a connector key. It is handled exactly as described below, except that no page preview is sent back: the answer lists the findings read, with the page and the words each came from, and nothing is estimated until the findings are sent for an estimate.
- What happens to the file. Our server opens it in an isolated converter, which turns each page (at most five) into a plain image and drops everything else in the file, including its metadata and photo location data. The original file is never passed on.
- What is sent, and to whom. The page images are sent to Microsoft Azure OpenAI (the gpt-6-luna model, run by Microsoft for us inside its US data zone), with our instructions to copy out the inspection findings and the unit facts printed on the page (bedrooms, bathrooms, size, year built, ZIP code). It is told not to copy names, addresses, phone numbers or signatures, and we remove any that slip through. The rows read from the page are then sent as text, with contact details removed, to the same model, as the AI reader does. Prices are always worked out by our own rules, never by the AI model.
- What we keep. Nothing from the file. The file, the page images and the text read from them are held in memory only while your upload is processed, then discarded; they are not written to storage or logs. A small preview of each page is sent back to your browser so you can check the findings against it. For a short time (up to a day) we keep the answer for a page in memory, found by a one-way hash of the page image, so the same page need not be sent to the model again. The findings you keep after checking them become part of your estimate and are handled like any finding you type, as described above. Our usage log line for each upload holds counts only (file types, number of pages, findings returned, timing, model and software versions) and the anonymous daily fingerprint.
- Show me where. In the page viewer you can ask where each item is on a page. Only then, that page's preview and the items read from it are sent to Microsoft Azure OpenAI again, and the answer is only boxes on the page. It is kept in memory for up to a day so the same page is never sent twice, and is not written to storage or logs.
- Limits. Each upload counts as one of your free daily calls, and there is a separate daily limit on uploads.
- What Microsoft does with it. The same as for the AI reader: Microsoft processes the images under its Azure OpenAI terms as our processor, does not use them to train AI models and does not share them with OpenAI. If Microsoft's automated abuse monitoring flags a request, Microsoft may store it in the United States so that authorized Microsoft employees can review it.
- Please redact personal details. Inspection reports often carry tenant names, signatures and addresses. They are not needed; cover, crop or otherwise redact them before you upload.
Pilot applications
The pilot page has a form to apply to our property manager pilot. It is optional, and it is the one place on the site where we ask who you are.
- What we store. What you enter: your name, work email, company, role, number of doors, state, move-outs a month, how you price turnovers today, your property management software and anything you write about move-outs. Likely names, phone numbers and other contact details in that free text are removed before it is stored, so please do not write about tenants there. We also keep that you agreed to be emailed about the pilot, the campaign tag of the link you followed (
ref), the time and the daily fingerprint (for spam control). - Why. To choose the pilot's agencies, to contact you about the pilot and to run it. We email you only about the pilot. We do not use your details for other marketing, and we never sell or share them.
- Who sees it. The application sits in a private Azure Storage container. A copy is emailed to our team inbox, and we email you a confirmation; both go through Resend, our email delivery provider in the United States. Applications are never posted to our issue tracker.
- How long. Up to 3 years, then automatically deleted. To withdraw, or to have your application deleted sooner (including the copy in our inbox), email us from the address you applied with.
Unsubscribing
We send short emails about the pilot to property managers one at a time, from our founder's own address. Each of those emails ends with how to stop them: the unsubscribe page, where you enter your email address, or a reply saying "stop". (Emails to an agency that has joined the pilot are described under Pilot agencies.) Outreach emails are sent from our Gmail account through Purelymail, and your address and any reply stay in that mailbox like any other email.
- What we store. A one-way hash (SHA-256) of the address, how it reached us (the page, a reply, a bounced email, or added by us by hand, for example after you tell us by phone) and the time. The list does not hold the address itself: the hash cannot be turned back into it, but anyone who already has the address can compute the same hash and see whether it is on the list. That is how our own check works: before we send any outreach email, we check the address against the list, and we do not email an address that is on it.
- Who sees it. The list sits in the same private Azure Storage container as pilot applications. The page gives the same answer for any address, so it does not reveal whether we ever wrote to you.
- How long. For as long as we send outreach emails. We do not delete it on a timer, because it is what stops us from emailing you. If you ask us to delete it, we will, but then nothing stops a future email.
Pilot agencies
If your agency is accepted into the pilot, we send its contact a pilot link for the agency. The pilot runs on that link, not on accounts.
- The pilot link and its cookies. The link lets your team, and an AI assistant you connect with it, use every tool free during the pilot, and it tags what you do with your agency. Opening it in a browser sets the
tk_pilotcookie, and opening the exit interview link sets thetk_interviewcookie, both described above. They carry your agency's pilot code, which is a short form of your company's name with a few random characters, and never a person's name or email. Please do not share the link outside your agency. - What we store, tagged with your agency.
- Each turnover estimate you run: the same facts as the anonymized estimate record (ZIP3, unit facts, standards, finding codes, totals), plus the estimated cost behind each finding code, in whole dollars (the 60 largest if there are more), which tool and channel, the time to the second and the estimate's hashes. The anonymized estimate record itself also carries your agency's tag.
- Each deposit settlement you run: the state, how many deductions there were and of which kinds, how many were allowed, the amounts (deposit, claimed, allowed, interest, refund due, tenant owes) and the return deadline. We never keep the deduction descriptions.
- Each document you have the turnover tool read (counts and status only, never the file), and each time you tick "Don't keep my notes" (only that you did, not what you typed).
- The page views, reading signals and usage log lines from your browser or assistant, kept 30 days like everyone's.
- The correction log: the whole text of your notes, or the list of issues your assistant sends, and for each finding the words it came from, what our matcher, the AI reader or a document reading proposed, and what you kept, changed, removed or added. We remove contact details, addresses, unit numbers, access codes and likely names before it is stored. It also carries the daily fingerprint and a network tag (a keyed one-way hash of the block of 256 IP addresses you connect from, which cannot be turned back into an address). It names the estimate the findings went into and, for a finding that was already in an earlier estimate in the same visit on the web page, the first estimate it was in, so we can compare what that line cost before and after a change. We keep it for pilot agencies even while we do not keep other people's notes. If you tick "Don't keep my notes" on the web, or send
"retain": falsethrough the API or an assistant, we keep none of it. - The ratings and feedback you send from a result.
- Your answers to our follow-up questions: what a job actually cost (in total, and per line if you give it), whether the tenant disputed it, whether the findings became deposit deductions, and a note. Free text is masked as above.
- Invoices you choose to upload with an actual cost: up to three PDFs or photos per job, 10 MB each. We do not mask them. Before a photo is stored we remove the hidden tags phones and cameras put in it, such as where and when it was taken and the device used; we keep only which way up it is, and the picture itself is not changed. A PDF is stored exactly as you send it, including anything hidden inside it. The one exception: if we find a hidden location tag in a PDF (for example, where a photo inside it was taken), or we cannot check the PDF, we do not store it and we ask you for a different file. This check does not read the invoice itself. We cannot promise to find every hidden detail in a PDF. An invoice may show a property address, a vendor, amounts and a tenant's name if it is not covered. Please cover tenant details before you upload. Uploading is optional.
- Your answers to the baseline question and the optional weekly questions (minutes per estimate, the biggest snag, anything else), masked as above.
- The exit interview at the end of the pilot: a typed conversation with our AI interviewer. It says it is an AI, and Corey Kozak reads every transcript. Each message is masked as above before it is stored or sent to the model. We keep the transcript, the language it was held in (first taken from your browser's language setting, then from the language you write in), the facts from your pilot data it was given to ask about (such as what a line cost before and after you changed it), a summary of your answers (including what you would pay and whether you would sign up), and up to three lines from your answers that we would like to quote, each with your yes or no and whether you want to be named or anonymous. We publish a quote only if you say yes to it at the end of the interview.
- Your agency's record: the company name, the contact's name and work email (from your application, or as you gave them to us), states, pilot dates and status. For each job and week we ask about, we also keep when we emailed you and a short description of the job (ZIP3, unit size and our estimate).
- Why. To run the pilot, to measure whether our estimates match real costs and save you time, to fix what is wrong, and to decide what to build and charge for. We do not sell or share it, and we do not use it for marketing outside the pilot.
- Emails. We send the welcome email with your pilot link, the follow-up questions and the weekly check-in through Resend, our email delivery provider in the United States. They go to your agency's contact, with a copy to our team inbox, and Resend keeps its own copy under its own terms. A follow-up email names each job by its date, ZIP3, unit size and our estimate, and links to a short form. Answering is optional.
- AI processing. The exit interview's masked messages, your company's name and the masked facts from your pilot data it asks about are processed by Microsoft Azure OpenAI in Microsoft's US data zone, under the same terms as the AI reader.
- How long. We delete each pilot record automatically 3 years after it was last written. The correction log is kept up to 5 years, estimate records and ratings up to 730 days, and log lines 30 days. Because everything is tagged with your agency, we can find your pilot records. To get a copy, to have them deleted sooner or to leave the pilot, email support@tektolo.com from your agency's contact address. We do this by hand. A deleted record can still be recovered from our storage for at least 30 days before it is gone, log lines expire on their own after 30 days, and we cannot delete the copies Resend keeps of our emails.
AI assistants
If you use the Service through an AI assistant (for example via MCP), the assistant's provider receives what you tell it under that provider's own policy. We receive only the structured request the assistant sends us, and handle it as described here.
Where data is processed
Our processor is Microsoft Azure, which hosts the Service and stores logs and estimate records in data centers in the United States, and runs the AI reader and the pilot's exit interviewer (Azure OpenAI, which processes the text you send it inside Microsoft's US data zone, and reads the page images of documents you upload). Microsoft handles that data under its own terms for hosting customers. Paid calls are also processed by the payment facilitator described above, and pilot applications and pilot emails by Resend (see Pilot applications and Pilot agencies). We do not otherwise share the data with third parties. Estimate, feedback, paid-call, notes and pilot records sit in private Azure Storage containers with no public access. The Service's managed identity writes them. Our administrator account can read them, and writes pilot agencies' records and the relayed feedback copies described under Feedback. Our deployment pipeline's identity can only read them, to prepare the daily feedback summary. A separate identity that sends the pilot's follow-up emails can read and write the pilot records and nothing else.
Security
All traffic uses HTTPS. We use private storage with identity-based access, no stored shared keys, a body-size cap and rate limits, and we keep personal data out of our records by design. No system is perfectly secure; if we learn of a breach affecting personal information we will notify those affected as the law requires.
Children
The Service is for property professionals and is not directed to children under 13, and we do not knowingly collect personal information from them. If you believe a child has sent us personal information, contact us and we will delete it.
Your privacy rights
Depending on where you live (for example California and other U.S. states with privacy laws), you may have rights to know, access, correct or delete personal information, and to opt out of its sale or sharing. We do not sell or share personal information. Based on current size we expect to fall below thresholds that trigger those laws, but we will respond to reasonable requests anyway where we can.
The practical limit: our estimate records contain no name, contact detail, IP or address, and the fingerprint rotates daily, so we generally cannot tell which records are yours and cannot access or delete them on request. If you can give us a specific detail (for example the date and what you typed into an unmatched finding), we will try to find and delete a matching record. Pilot records are the exception: they are tagged with your agency, so we can find, copy or delete them on request, by hand and with the limits described in Pilot agencies. We will not discriminate against you for making a request. To make a request, email support@tektolo.com. We may need to verify that a request is genuine.
Changes
We may update this policy. When we do, the effective date above changes, and for a material change we add a line under it saying what changed. The version in force when you use the Service applies. If we add accounts, other payment methods, analytics or cookies, we will update this policy before doing so.
Contact
Corey Kozak · support@tektolo.com · https://tektolo.com