Privacy Policy & AI Use Notice

DeskFamiliar is developed and operated by 长沙林讯科技有限公司 (the operator, “we” or “us”). This notice explains information handling across our website, desktop app and DeskFamiliar Cloud, together with the limits, permissions and choices involved in using generative AI. Please read it before using the relevant features, particularly the sections on request transmission, third-party processing and your rights.

What to know first

  • Session history is primarily stored on your computer. Cloud AI still receives the conversation, context and tool results needed for each request.
  • The DeskFamiliar Cloud gateway does not persist conversation or file bodies as content records. It retains necessary account, quota, transaction and security records; upstream model services have separate retention rules.
  • DeskFamiliar does not use your conversations to train models. Check the separate practices of custom endpoints, external Agents and third-party plugins.
  • AI answers, code and suggested actions can be wrong. Character animations indicate activity; verify permissions and actual results before relying on file, command or external-service operations.

1. Scope and service roles

This notice covers our website, desktop shell and built-in cloud AI service. Hermes Agent runs built-in conversations, tools, memory and sessions; DeskFamiliar supplies the desktop interface, 3D character, permission prompts and activity visualization. Third-party websites, model services, external Agents, Skills and plugins may also have their own terms.

DeskFamiliar is not affiliated with Nous Research or the official Hermes Agent team. References to Codex, Claude Code and other products describe compatibility, not partnership, affiliation or endorsement.

Reading this notice or visiting the website does not authorize all tools, file access or optional processing. Enable features through the relevant controls. This notice does not replace separate consent where required by law.

2. AI capabilities and risks

AI generates content from input and context. It can misunderstand instructions, invent facts, sources or code, and produce incomplete, outdated, biased or unsuitable results. Answers may vary. Model, network and tool failures can delay, interrupt or leave tasks unfinished.

Output assists understanding, creation and work. It does not replace qualified medical, legal, financial or other professional judgment, serve as an emergency contact, or provide a sole basis for consequential decisions. Before executing code, changing files or publishing results, check facts, operating conditions and effects, and back up important work.

The character’s expressions, language and animations are software presentation, not evidence of human consciousness, feelings or professional qualifications. A completion animation reflects a task event, not a guarantee of correctness or the intended external result. These cautions do not exclude our legal responsibilities or your statutory consumer and privacy rights.

3. Information we process

The information involved depends on the features you use. Local storage, transmission to complete a request and persistent cloud records are distinct.

  • Local data: conversations, session history, Agent memory, tool records, staged attachments, drafts, settings and runtime logs may be stored on your computer to resume sessions, run tasks and troubleshoot. This does not mean every local record is uploaded.
  • AI requests: your messages, relevant conversation history, memory or system context, file excerpts, attachments, images and tool output may form a model request. Image and other capabilities depend on the selected model and app version.
  • Steam account information: binding processes your numeric Steam ID (SteamID64) and authentication ticket. The service also handles Steam-provided display name, avatar URL, country or region, wallet currency and relevant account status for display, verification, regional pricing and transactions. DeskFamiliar does not ask for your Steam password.
  • Usage and transactions: model and request identifiers, input and output token counts, billed quota, request time and outcome, quota grants and deductions, order and transaction IDs, amount, currency, and purchase or refund status support metering, delivery, reconciliation and disputes.
  • Security and technical information: requests expose information such as IP address, time and outcome. Authentication and security records may also include client version, device identifiers where supplied by a client, and error categories for abuse prevention, rate limits, session security and troubleshooting.
  • Support submissions: contacting us provides your email, issue description and any request IDs, order information, screenshots or logs you choose to share. Remove unrelated personal information, credentials and file contents first.

4. Purposes and necessity

We process information as needed to provide requested features: verify identity, generate responses, run enabled tools, restore local sessions, display character activity, allocate and settle quota, deliver purchases, handle refunds and support, and maintain security. Personal information is processed on a basis permitted by applicable law, such as necessity to provide the relevant service, consent or a legal obligation.

Without using cloud AI, that feature does not send model requests. Withholding information necessary for a task may prevent its completion. Declining optional features or unnecessary processing should not affect unrelated basic features. This notice does not grant general access to your entire disk or authorize unrelated future uses.

5. Built-in models, provider processing and filings

Built-in requests follow this path: local Hermes Agent → DeskFamiliar Cloud → Alibaba Cloud model service → response. Current models are Alibaba Cloud’s Qwen (Tongyi Qianwen) qwen3.7-flash, qwen3.7-plus and qwen3.8-max, with inference in the Beijing region of mainland China. Available models appear in the app. We will update this notice when the provider or processing changes.

Alibaba Cloud receives the request content and model parameters needed to respond. Our gateway does not persist conversation, attachment or tool-returned file bodies as content records, but it still relays them for upstream processing. Alibaba Cloud’s official notice states that relevant data is not used for model training and that model and application call data is stored as required by law. See its official notice for processing, retention and security terms; this is not an end-to-end zero-retention promise.

Alibaba Cloud lists Tongyi Qianwen under generative AI filing ZheJiang-TongYiQianWen-20230901 and the algorithm named “达摩院交互式多能型合成算法” under 网信算备330110507206401230035号. These identify the upstream model and algorithm, not a DeskFamiliar filing or application registration, and do not guarantee output quality. Check official notices for current records.

6. Custom endpoints, local models and external Agents

In versions that support custom endpoints, selecting one sends model requests directly from your computer to the configured service, bypassing DeskFamiliar Cloud without uploading that endpoint’s model usage to it. API keys are stored in the operating system credential store. Check the address, provider policy and processing region. Continuing an existing session after switching providers can include its earlier context in subsequent requests.

A local model means the inference endpoint is local. Search, browser, MCP, Skills and other network tools may still send task information to their services; a local model does not make every feature offline.

Built-in Codex and Claude Code observers read relevant local session activity to animate the character. They do not send external Agent conversations to DeskFamiliar Cloud or take over chat, approvals or permissions. The external Agent’s own data handling depends on its configuration and provider.

7. Tool permissions and third-party plugins

File, command and web tools process task information, and results may enter later model context. A read-only operation can still supply content for cloud inference; read-only does not mean no transmission. Use directories you are authorized to access and check the information involved before sending or approving.

The built-in Agent uses read-first defaults, risk-based confirmations and blocks for certain dangerous commands and sensitive paths. Persistent permissions are auditable and revocable in security settings. Whether a prompt appears depends on the operation, settings and existing permissions; do not assume every read, write or network request asks separately. Disable unused tools and connections through their controls.

Third-party bridge plugins run as local programs with your user privileges, not in a sandbox. They may read or write files and access networks independently. Confirmation at first run or after entry-point changes is not an endorsement of their behavior. Install trusted plugins and review their authors’ data notices. Manage external Agent permissions in the relevant Agent.

8. Training and sensitive information

DeskFamiliar does not train foundation models or use your conversations or files to train models. Processing needed to answer a request is not automatically permission for product analytics, human evaluation or model improvement. If such optional uses are introduced, we will explain them and obtain legally required consent first; declining will not affect the relevant core features. Check custom services and external Agents separately.

Avoid unnecessary identity documents, financial accounts, health details, precise location, children’s information, passwords, API keys, trade secrets or other people’s private information. Redact where possible; files read by tools may also contain this information. Ensure you have a lawful basis to supply materials and meet applicable notice, authorization or separate-consent requirements for sensitive or third-party personal information.

9. Retention and deletion limits

Local sessions, memory, staged attachments and logs occupy different locations and remain according to the feature and your cleanup actions. Deleting a session may not remove independent memory, project files, logs or backups; uninstalling should not be treated as complete erasure. Check what work you need to retain and contact us for help locating relevant data.

Cloud account, quota, transaction and security records are retained as needed for service, reconciliation, disputes and legal obligations. Duration depends on the record type, account status, unsettled transactions and applicable requirements. We do not currently promise a single automatic deletion interval or immediate erasure on sign-out. Request a review or deletion under section 16; where records must remain, we will explain the basis and scope and limit their use accordingly.

Local deletion or stopping use does not automatically erase copies lawfully processed or retained by third parties. Model providers, Steam and services you choose have separate retention rules. We will assist with requests related to our service where needed.

10. Sharing, disclosure and regions

Relevant services receive information necessary for their functions: Alibaba Cloud processes model requests; Steam (Valve) verifies accounts and handles wallet payments and orders; enabled network tools or third-party services receive data needed for their tasks. Website hosting and network infrastructure also process connection information. We do not sell your conversations or personal information.

Beyond these functional needs, your authorization or legal requirements, we do not use personal information for unrelated third-party purposes. For entrusted processing or external provision, we define purposes and scope and meet applicable notice, contractual and consent obligations. Legally required disclosures are limited to necessary information.

Built-in inference currently takes place in Beijing, mainland China; use elsewhere transmits requests to that region. Steam, website infrastructure and custom providers may process information in other regions, so we do not promise that all data always stays on your device or within China. Transfers subject to legal restrictions require applicable procedures; this notice does not replace separately required notice and consent.

11. Information security

DeskFamiliar Cloud uses HTTPS, credential verification, access controls and security auditing. The client stores relevant API credentials in the operating system credential store. These measures cannot eliminate all risk or imply that every local session file, log or backup is encrypted. Protect your device account, operating system and credentials, and grant tool and plugin permissions carefully.

If you suspect a data leak, unauthorized access or exposed credential, stop the affected operation, revoke relevant permissions and contact us below. We will investigate, take appropriate remedial action and meet applicable notification or reporting duties. Do not post keys, complete logs or private files in public issue trackers.

12. Website, cookies and external links

The current website is a static information site. Its code sets no cookies and includes no behavioral analytics or advertising trackers. Browsing still sends requests to hosting and network infrastructure, which may process IP addresses, browser request information, access times and errors to serve and protect the site. Static hosting does not mean no network data processing.

Following links to Steam, GitHub, provider policies or other sites brings you under their cookie, account and logging practices. Read their notices. If we introduce analytics or other unnecessary processing, we will update this explanation and provide choices as required.

13. AI labels, conduct and intellectual property

The current app marks AI replies with “AI 生成” and displays “内容由 AI 生成,请仔细甄别” below the composer when using Simplified Chinese and DeskFamiliar Cloud. Character actions, presentation or the absence of that badge do not prove human authorship. When copying, exporting or publishing, check and retain or add AI disclosures as applicable; do not maliciously remove, alter, forge or conceal legally required labels.

Do not use the service for unlawful or fraudulent content, privacy or intellectual-property violations, discrimination, harassment, attacks, or bypassing permissions, security or metering. Models and tools may reject unsupported or restricted requests; this does not mean every violation can be detected.

Ensure you may lawfully submit and use your materials. Output can resemble existing works or contain unsuitable content. This notice does not guarantee originality, copyright eligibility or freedom from third-party claims. Check references, permissions and facts before commercial use or publication. Your responsibilities do not remove ours under applicable law.

14. Children and minors

Minors should read this notice, select features and use AI with a guardian’s guidance. A virtual character should not replace real relationships or qualified help. Purchases require appropriate guardian consent and compliance with Steam account and payment rules.

Do not submit unnecessary personal information about minors. Under Chinese law, personal information of children under fourteen is sensitive and requires special protection and appropriate guardian consent. Guardians who believe information was improperly submitted or a purchase lacked appropriate consent can contact us; we will investigate and handle the matter under applicable law or assist with the payment platform.

15. Quota, payments and disputes

Built-in AI offers a Steam-linked free monthly quota and one-time quota packs through Steam Wallet, without a subscription. Free monthly quota refreshes each period and does not roll over; purchased quota has no expiry date. Available quota, packages, prices and model charging rules are shown in the app at purchase and use.

A task can involve multiple model calls, historical context and tool results. Billed usage is not the same as the number of characters typed or received; model, caching and billing rules can affect deductions. Whether a failed or interrupted request incurred usage requires checking actual generation and settlement records.

Steam handles payment. We do not receive complete bank-card details or Steam payment passwords, but retain transaction information needed to deliver and reconcile purchases. Purchases, refunds and disputes follow applicable law and Steam rules; this notice creates no blanket no-refund exclusion of statutory rights. Send an order or request ID and time if you suspect duplicate deductions, missing quota or incorrect usage.

16. Your rights and contact details

Subject to applicable law, you may request information about processing, access, copies, correction or deletion, withdraw consent-based authorization, request processing to stop, or request closure of account data associated with this service. We will explain why transaction, security or legally required records must remain. Stopping use or withdrawing consent does not invalidate earlier lawful processing.

Manage sessions, memory, tool permissions, endpoints and external Agent connections in the app. Disabling a connection stops future observation; revoking a tool permission limits future operations. These actions do not automatically erase existing cloud records. Exercise Steam and other provider account rights through those platforms.

Operator: 长沙林讯科技有限公司. Contact support@deskfamiliar.com for privacy requests, AI-content complaints and service support. Identify the feature and your issue or request; for account records, provide necessary references such as your numeric Steam ID, request ID or order ID. Do not send passwords, API keys or unredacted complete files.

We may verify identity proportionately to the request to prevent impersonation, and will handle and respond within applicable legal time limits. If more information is needed, a request cannot be fulfilled or records must remain, we will explain why. You retain applicable rights to contact competent authorities or seek other remedies.

17. Updates to this notice

We will update this notice as features, model providers, processing practices or applicable requirements change, and show the revision date here. Material changes to purposes, scope, recipients or rights will receive an appropriate notice. Where renewed consent is required, the relevant processing will begin after obtaining it.

Continued browsing is not a substitute for specific authorization, and updates do not retroactively broaden existing consent. Contact us at the email above with questions about this notice or current processing.