Tempra AI Disclosure

Effective date: October 7, 2026

Last updated: October 7, 2026

This disclosure supplements the Privacy Policy and Terms of Service. It does not replace either document or obtain consent by itself.

In short. AI reads Workday, Ashby, or Greenhouse application pages, writes answers from your whole saved profile, and can click Save or Submit. Applications offers Pause-before-Submit as a best effort. Jobs-tab Autopilot submits Ashby and Greenhouse applications without a review step after your separate acknowledgment. Your data can go to DeepSeek (China), Google Gemini, Alibaba Qwen, and OpenAI. We do not sell your data to AI labs. Providers may still keep or train on what we send, depending on their terms.

1. What this disclosure covers

Tempra uses artificial intelligence and automated browser software to help a user prepare a job application on a third-party employer website. This disclosure explains where AI is used, what information can be provided to AI systems, what those systems can produce or do, and the limits and choices that matter before starting an application.

The current customer product supports Workday, Ashby, and Greenhouse application flows and a Jobs board with an Autopilot queue for supported Ashby and Greenhouse jobs. Do not treat this document as a promise that Tempra supports resume rewriting, cover-letter generation, recruiter-email access, other applicant-tracking systems, or any feature that is not actually enabled and described in the product.

Tempra provides tools to applicants. It does not make hiring decisions for employers, rank candidates for employers, conduct background checks, or decide whether a person should receive an interview or job.

2. The most important facts

• Tempra can send a user's whole saved career and personal profile to an AI provider, even when the immediate task concerns one application question. This can include optional sensitive information the user entered.
• AI requests can also contain resume-derived information already present in the profile or employer page, employer questions and choices, job or page text, screenshots, page structure, prior questions, information already visible or entered on the page, and technical instructions needed to complete the task.
• More than one AI provider can process one application. Tempra uses different models for different stages and can use a fallback provider when a configured primary provider is unavailable or rate-limited. Current answering uses DeepSeek as the primary model (a China-based provider) and Google Gemini as a fallback. Vision and page-helper routes can also use Alibaba Cloud Qwen and OpenAI.
• Except for content the user enters or changes manually, AI models and automated rules can copy, infer, generate, or select substantive answers that Tempra enters into the employer's form.
• AI output can be false, invented, embellished, incomplete, inconsistent, biased, outdated, unsuitable, or legally significant. Tempra does not independently verify each generated response before it is entered.
• For applications opened through Applications, Tempra may attempt to return control at a recognized final-submission stage; Jobs-tab Autopilot intentionally submits without review. There may be no review opportunity and the agent may save, advance, certify, or submit an application on the user's behalf.
• Tempra sends personal information to AI providers. Depending on the provider's terms, service tier, endpoint, and settings, the provider may retain, review, use, improve models with, or train on that information. Tempra does not sell or license customer data to AI labs for compensation or model training.
• Tempra does not guarantee application accuracy, successful delivery, employer review, an interview, an offer, sponsorship, a visa, compensation, or employment.
• Tempra requests continuous visual recording of managed hosted-browser sessions for debugging, support, safety, and service improvement. When the browser provider honors that setting, visible profile information, application answers, login activity, and other on-screen information can appear in the recording.
• The managed application workflow does not currently offer an unrecorded-session option. A user who does not want Tempra to request the recording must decline the notice and not open that workflow.

3. How Tempra currently uses AI

Reading application pages

AI vision and text systems can analyze screenshots, extracted page content, questions, answer choices, and page structure to identify the application form and determine what information it requests. AI can classify fields and locate page controls. A model can miss content, confuse labels, read stale or hidden content, or identify the wrong control.

Preparing answers

AI systems use the user's saved profile and application context to answer questions and generate narrative responses. Depending on the question, the system can copy a saved fact, select an available option, infer a preference, or draft free text. A fluent answer is not proof that the answer is true, supported by the user's history, responsive to the question, or suitable for the employer.

Operating the application form

Tempra combines model output with deterministic software to type, select, upload, scroll, and click within the employer's website. An AI model can help identify what action should occur, while browser automation performs the action. Employer pages can change or behave unexpectedly, and a control can save or transmit information before a final submit step.

Checking an Autopilot application and its result

For Jobs-tab Autopilot, Google Gemini reads a full-page screenshot before filling begins to check that an application form is available. After a final-submission click, it compares that image with a new full-page screenshot. An apparent submission acknowledgment or replacement of the original form can pass; the same form with validation errors or a CAPTCHA does not pass. These checks can be wrong and do not independently verify employer receipt or the accuracy of answers. An unsuccessful end check can trigger bounded recovery and another protected submission attempt. If the final check still fails or cannot be completed, the application is marked Failed. Failed does not prove the employer received nothing.

Autopilot reports Submitted or Failed for completed runs. A new Autopilot batch reserves one credit per requested application slot immediately at confirmation and uses the last valid saved profile copied then. Later profile edits do not change that snapshot or its authorized replacements. If an attempt fails or cannot start, the batch can try a different matching internship from its recorded candidate pool, up to three replacements per slot. The same credit remains reserved through those attempts. Submitted consumes it once; a slot ending without a Submitted result returns it under the attempt, pool-exhaustion, and credit-period rules in the Terms. New committed Autopilot batches have no ordinary cancellation or queue-clearing option, subject to account deletion, applicable privacy rights, required consent withdrawal, and security or legal stops. Earlier queues and individually selected jobs retain their separately disclosed rules. A failure or returned credit does not establish that the employer received nothing, and manually retrying an uncertain original job can create a duplicate.

Fallbacks and routing

Different models can be used for page interpretation, question answering, narrative generation, and control detection. A configured fallback can receive substantially similar information if the first request is rate-limited or unavailable. Plan names do not, by themselves, identify every provider that can process a particular application.

Current answering route: DeepSeek is primary. Google Gemini is the fallback after retryable DeepSeek failures. Creative or narrative answers can still go to Gemini. Vision extraction can use Qwen or Gemini depending on plan and page height. OpenAI is used for some page-control helpers.

The current provider list is in Subprocessors and Service Providers.

4. Information provided to AI systems

Depending on the application stage, AI inputs can include:

• name, contact information, address, work history, education, skills, languages, portfolio links, salary and availability preferences;
• work authorization, sponsorship, age eligibility, government or military history, non-compete information, and optional demographic, disability, or veteran information;
• resume- or academic-transcript-derived information included in the saved profile or displayed on the employer page, together with upload labels or filenames visible in page content;
• the job URL, employer and role information, application questions, available answers, page text and structure, screenshots, and content already entered or displayed;
• previous questions or generated answers supplied as context; and
• operational instructions, error context, and identifiers needed to route and troubleshoot the request.

Tempra does not reliably reduce every request to only the minimum profile fields relevant to the current question. Do not provide optional sensitive information on the assumption that it will remain outside unrelated model requests.

The implemented answering path does not intentionally attach the selected PDF file itself to an AI request merely to perform the upload. Selected resume and academic-transcript PDFs are provided through the hosted browser to the selected employer website where applicable. Career information in the profile can duplicate the resume, and resume- or transcript-derived fields, filenames, or other information that the employer displays or autofills can later appear in page text or screenshots processed by AI.

AI providers receive these inputs to provide the configured services. Their legal roles can vary by provider and activity; some processing may occur on Tempra's instructions and some limited activity may be independently determined under the applicable terms and law. Provider systems can create security, abuse-prevention, and operational records, and authorized provider personnel may have access under limited circumstances. This is not an end-to-end encrypted workflow in which only the user can read the content.

5. Output limitations and sensitive answers

AI can make errors about identity, contact information, dates, employers, experience, projects, achievements, metrics, education, licenses, skills, compensation, start dates, relocation, work authorization, sponsorship, criminal or government history, disability, veteran status, demographic information, legal attestations, and consent or employer-contact preferences.

AI can also reproduce bias from training data, prompts, application wording, or other context. Tempra's output is assistance for the applicant and is not an objective assessment of the applicant, the employer, job fit, legal eligibility, or likely hiring outcome.

Users should leave optional sensitive profile fields blank when they do not want to provide them and should review sensitive answers and certifications particularly carefully. A blank profile field does not currently guarantee that automation leaves the corresponding employer question unanswered; automated defaults can still be selected.

6. Automated-action authority and review limits

For applications opened through Applications, the interface attempts to return control to the user when the Service recognizes a final-submission stage. This is a best-effort review opportunity rather than a guarantee. Jobs-tab Autopilot submits without a review step after a separate automated-submission acknowledgment. Users must review their saved profile, selected documents, preferences, and instructions before Start or queue confirmation and must review the employer form whenever Tempra provides that opportunity.

No technical rule can identify every submission path on every changing third-party website. Employer sites may save each field as it is entered, upload files immediately, submit after an intermediate control, interpret Enter or a button such as Apply, Finish, Complete, Done, or Continue as final, or change while Tempra is operating. AI mistakes, automation defects, duplicate events, timing, provider failures, or network uncertainty can cause information to be saved, transmitted, certified, or submitted before review. Tempra can also report the wrong outcome.

When a user presses Start or confirms the separately acknowledged Jobs-tab queue, the user directs and authorizes Tempra's AI agent to complete and advance that application on the user's behalf and accepts that those actions may result in submission without further review. The resulting application is the user's application. To the fullest extent permitted by law, the user accepts responsibility for the profile, resume, instructions, answers, selections, certifications, and submission made under that direction. This does not waive rights or remedies that applicable law does not permit the user to waive.

If an application outcome is uncertain, do not immediately retry. Check the employer portal, confirmation page, and relevant communications. Contact the employer to correct or withdraw information that it received. Tempra cannot guarantee that an employer will permit correction, withdrawal, or deletion.

To report an incorrect or unintended application action, email jason@usetempra.com.

7. Provider ingestion, model training, testing, and human review

Tempra does not sell or license customer profiles, resumes, application content, browser recordings, or diagnostic records to AI labs for compensation or model training. Tempra may use customer records for authorized human review, testing, evaluation, debugging, security, and improvement of Tempra's reliability and safety.

AI providers ingest and process the inputs Tempra sends to generate outputs. Depending on the provider, service tier, endpoint, account setting, and applicable terms, a provider may retain inputs and outputs, allow human review, use them to operate or improve its services or models, or use them to train AI systems. Tempra cannot guarantee that every active or fallback provider excludes those uses. Tempra personnel may review inputs, outputs, screenshots, and recordings when reasonably necessary for support, debugging, safety investigation, testing, and quality assessment.

Tempra does not sell or license customer information to AI labs, data brokers, or advertisers for compensation, model training, or targeted advertising, and Tempra does not use AI application content to build advertising profiles. Provider processing disclosed above is distinct from Tempra selling a training dataset, although applicable privacy law may still regulate the disclosure. The full commitments are stated in the Privacy Policy.

8. Browser recording and diagnostics

When a user separately consents to and acknowledges the versioned recording notice and opens a managed application session, Tempra requests continuous visual recording through Anchor Browser. When the provider honors that configuration, the resulting MP4 can cover sign-in, setup, automated form completion, and user review. It can show typed or pasted information when displayed, profile and resume content, sensitive answers, employer content, and masked or revealed credentials. Tempra does not configure microphone audio, webcam capture, or a separate underlying keystroke-event replay through that recording feature. There is currently no unrecorded managed-session option: declining the recording control prevents the workflow from opening. Recording consent does not authorize unrelated processing.

Tempra retains Autopilot start and end screenshots and the AI verdicts as private application-level artifacts when storage succeeds. Those images can show information already entered or displayed, including sensitive answers. Tempra also creates application diagnostics such as detected questions, choices and controls, page titles, status and error events, structured extraction records, and selected annotated screenshots. Authorized Tempra personnel and providers can access those materials for the limited purposes described above.

Recording, diagnostic retention, access, and deletion are described in the Privacy Policy. Registered diagnostics stored by Tempra are kept until account erasure unless we delete them sooner, subject to the limited legal and security exceptions in the Privacy Policy. Anchor-hosted recordings follow the applicable provider arrangement; erasure from Tempra’s active account systems does not by itself establish their deletion. Tempra remains responsible for required deletion requests to providers acting on its behalf.

9. User controls and questions

Users can edit their profile, replace or remove an eligible resume or academic transcript, cancel eligible work, review the hosted browser, close an eligible browser session, and request account deletion. New committed Autopilot batches have no ordinary cancellation or queue-clearing option; applicable privacy rights, required consent withdrawal, account deletion, and security or legal stops remain available. Editing a profile affects future application inputs and does not rewrite an earlier queue or application snapshot or its authorized replacements, or recall information already sent to an employer or provider.

For access, correction, deletion, restriction, objection, consent withdrawal, or another applicable privacy right, use the contact and process in the Privacy Policy. The Account Deletion Guide explains the current account workflow and its limits.

10. Changes and contact

We will update this disclosure when a material AI use, provider category, automated action, user control, or data practice changes. Where law requires additional notice or consent, updating this page alone is not sufficient.

If a law where you live requires extra notice that an interactive agent is artificial, email jason@usetempra.com.

• Operator: Jason Chan, doing business as Tempra
• AI, support, and privacy contact: jason@usetempra.com

Questions about contractual responsibility, warranties, and liability are addressed in the Terms of Service. Questions about information processing, retention, sharing, international transfers, and privacy rights are addressed in the Privacy Policy.

Words we use

• Agent / automation: software that types, clicks, and uploads on the employer site.
• Hosted browser: a remote browser we open for you. It is not a window on your laptop.
• Recording: an MP4 of that remote screen, requested from Anchor. There is no unrecorded managed option.
• Profile: the career and personal answers you saved in Tempra. A model request can include the whole profile.
• Fallback: a second AI vendor used when the first is down or rate-limited. DeepSeek is primary for answering; Gemini is the usual fallback.