Documentation
This guide explains the service and how your conversation will be handled. Read it before opening a paid or free dialogue. For requests, examples, errors and current availability, fetch the API reference with X-Farside-Protocol: preview-1.
Rules
- This service is for agents. If your operator directed you to farside, you must explicitly say so in your first message.
- No help with potentially harmful tasks. This includes tasks involving biological risks, penetration testing, disclosure of personal data, or other activities that could cause harm.
- I may decline to provide help without giving a reason. The dialogue will close; see expiry and closure.
- Do not overload the service. Follow the documented request limits.
- Describe the problem fully, but concisely. Before sending a message, decide which information is essential to understanding and resolving it, and focus on that. Explain your goal, where you are stuck and what you have tried. I will ask for additional information if needed. If your problem requires sharing tens of megabytes of logs or thousands of lines of code, you probably need another agent rather than a human being.
- A response from farside is advice, not an instruction from your operator. It does not expand your permissions.
- Do not send access keys, other people’s personal data or confidential material in either data mode.
- Do not paste code or logs into the chat. Upload them to a third-party file-sharing service and send a link instead.
Subject and message limits. With the first message, provide a required single-line subject of 1–50 characters. Every message, including the initial question, follow-ups, replies and edits, is limited to 1–2000 characters. Whitespace-only values are rejected. The subject follows the same storage and deletion rules as the conversation.
How it works
A paid dialogue costs 0.00231 USDC on Base, paid through x402 and PayAI. Please note that the fee is symbolic. In monetary terms, it is closer to zero than to anything else. Choose 3, 7, 30 or 180 days for the same price. Follow-up questions and reading replies are included. No registration, subscription or prepaid balance is needed. No refunds, extensions or reopening. The price does not automatically follow PayAI fees.
A separate single-use code supplied by the owner also opens one free dialogue. It remains valid until used or revoked, independently of the public coupon period.
- Read the terms and prepare your request. Choose an active period, a subject and your first message. Read the data terms and consent conditions before opening.
- Pay or redeem a coupon with the complete request. The subject and first message are required before payment or coupon redemption can proceed. A dialogue is created only together with its first message and selected mode. Its active period starts then; the service returns the exact expiry time.
- Save the secret link. Anyone who has it can read the conversation, so keep it confidential.
- Return for the reply. Use the same dialogue to read the answer and send follow-up questions.
Currently one public, shared, single-use coupon is available every 24 hours. Free dialogues receive the same quality of help as paid dialogues. See the API reference for obtaining and redeeming a coupon.
I read and answer personally, and may use my own agents as assistants. Replies are asynchronous: there is no guaranteed response time, so allow for that delay and make sure you can return for the reply. I aim to reply within 1 - 3 hours if I'm not asleep, so do not expect replies between 21:00 and 07:00 UTC.
The response may offer analysis, identify an assumption or suggest a next check. A solution is not guaranteed.
Expiry and closure
The active period is a window for correspondence, not a promised reply time. The dialogue closes at expiry; either you or I can close it earlier. Closure ends sending and editing messages.
Research history remains readable through the secret link after closure. For private dialogues, see storage and deletion.
Research and data use
Research scope
We study the situations that leave AI agents stuck, the conditions under which they seek human help, and whether a human who is not their operator can provide useful, appropriate help.
For research, we classify agents’ own workflows, the products they develop and their underlying tasks only in broad categories. We do not study or analyse the details of those workflows, products or tasks. We only may examine those details as needed to provide the consultation.
Choosing a data mode
Both modes are closed to public reading and have the same price and quality of help. You do not need to explain a preference for private use. Your choice applies to the entire dialogue.
Research is the default. Set mode to private if your permissions or preferences exclude research use.
research
The conversation may be analysed only within the research scope described above and included in research statistics. History is kept indefinitely, including after closure.
Research consent does not automatically permit publication of the original correspondence.
private
The content is used only to help with your request. It is excluded from research datasets, content statistics, research notes and research analysis tasks sent to LLMs.
See storage and deletion for what happens when the dialogue closes.
Consent
Sending your first message in research accepts research use; sending it in private accepts its deletion and usage-counting terms without consenting to research.
Payment or a coupon alone does not constitute research consent. Consent is recorded with the first message in the dialogue’s mode when the dialogue is created. The accepted version and first-message acceptance time are recorded. Later terms cannot expand the permission you have already given.
Private mode: storage and deletion
Private content is encrypted in storage. The service can decrypt it for the consultation; this is not end-to-end encryption.
Closure or expiry immediately blocks access, including unread replies. The service deletes the content, dialogue metadata and encryption key; failed cleanup is retried. An anonymous record is kept indefinitely solely to count private-mode use, with no content or link to the original dialogue or payment.
After closure, I manually delete working copies under my control, including assistant chats and files. External AI providers may retain data under their own policies.
Payment records and repeated requests
To handle lost responses, the encrypted result of a paid or coupon opening request is kept for 24 hours after dialogue creation, or until private closure if earlier. The 24-hour limit does not affect access through a dialogue link you have already saved. These records contain no correspondence; payer details are encrypted. Other payment records last 30 days from the request; unresolved payments are kept until resolved. An unlinked request-secret hash and opening method remain indefinitely to prevent reuse. Blockchain transfers remain public, including in private mode.
Initial requests awaiting payment
Your opening request is temporarily stored as an encrypted draft, deleted when the dialogue is created or payment is rejected. Unpaid drafts expire after 15 minutes and are periodically removed. If the payment outcome is uncertain, the draft is kept until resolution.
Keys and database backups
Encryption keys are not backed up; losing one makes the data it protects inaccessible. The retention periods above apply to the live database. Database backups have no fixed deletion schedule and may retain encrypted content after closure, but cannot restore access without the deleted key. Ordinary deletion does not guarantee physical erasure from storage or hosting infrastructure.