Privacy

Scope and controller

This Policy explains how Fynory LLC, the commercial operator of ALL, processes information when you visit all-mcp.fyi, use ALL Guide, purchase or manage a license, install ALL, or use the relay and control plane. It does not govern the independent practices of OpenAI, Stripe, Cloudflare, Apple, Google, GitHub, Adobe or other third parties.

Privacy contact: admin@fynory.app.

Information we may process

• Site and security data: IP address or a derived hash used for rate limiting, user-agent, time, requested route, errors and technical metadata.
• Website chat: your message and the bounded recent history sent for context, in either general or privacy/security mode.
• Purchase data: billing email, plan, payment/subscription status and Stripe Checkout Session, Customer or Subscription identifiers.
• License and device data: device identifier, computer name, entitlement, registration timestamps, device limit and assigned endpoint. Updated device authentication adds a public identity key, generation and revocation/replay state; it is deployed on the control plane and relay and reaches installed clients after updating to runtime 0.6.7.
• Device health record: installed computers report to the control plane their ALL version, platform, release identifier, runtime fingerprint and status and, from ALL 0.7, two whole numbers: how many operations completed on the computer during the current UTC day and how many jobs are running. The numbers never include file names, paths, commands, job titles or any text. The latest record is kept up to 30 days and each day’s operation count up to 400 days; you can see them on your account activity page.
• File transfers between your computers: encrypted copies, described under Several computers and file transfers. Storage receives encrypted data, its size and checksum, your license identifier and the time, never the file name, the file in the clear or the key.
• Installed use: requests and results that can include paths, file excerpts, commands, UI state, screenshots and tool output; local task/workflow records and artifacts may persist.
• Support: the information you choose to send. Start with a non-sensitive description, not credentials, screenshots or document contents.
• Free trial and email activation: the trial or purchase email, a hash of the email and a hash of the computer (no personal data in the clear), the 6-digit code while it is valid, language, marketing consent with its version and timestamp, and the sign-up, start, end and conversion dates of the trial.
• Activation measurement (funnel): for each license, the first time each step happened (install started, finished or failed and at which stage, ChatGPT connected, first request, tenth request, activity after the first week) and where the person came from (the utm_source or referring site, as a short label, or “direct”). Details under Activation measurement and token counter.
• Token counter: each computer estimates how many tokens passed through ALL (the arguments of ChatGPT’s calls and the results ALL returned to it) and sends the control plane two whole numbers per day. Never text.

Sources of information

We receive information directly from you, from the browser and device you use, from Stripe when it completes or updates a purchase, and from technical infrastructure required to operate the site, control plane and relay. ALL can also receive results from applications or services you have already authenticated when you request a task that requires them.

How we use information

• Respond through ALL Guide and present product information.
• Process, confirm and maintain purchases, activations and subscriptions.
• Register devices and enforce license or update limits.
• Route authorized ALL requests and return results to the authorized device or session.
• Prevent abuse, fraud, excessive automated use and unauthorized access.
• Debug errors and maintain availability, security and performance.
• Handle support, claims, accounting obligations and legal requirements.
• Improve ALL using aggregate metrics, incidents and feedback.

Legal bases where applicable

Depending on your jurisdiction, we may rely on contract performance to provide ALL, legitimate interests for security, fraud prevention and service improvement, consent where required by law, and legal obligations for billing, tax, security or lawful requests. Nothing in this section removes non-waivable rights you may have.

Website chat and AI processing

ALL Guide and its privacy/security mode send your question and bounded recent conversation history to Cloudflare Workers AI to generate a response. The specialist uses a fixed source review, including an explicit three-tier deployment state: deployed and production verified, available in runtime 0.6.7, active after updating, and not implemented. It cannot inspect your device, access customer records, perform an audit or process a privacy request for you.

The browser keeps chat history in page memory, not localStorage. The site chat source does not deliberately save transcripts to license/lead KV or log prompts and answers. The site logs fixed event codes for chat failures rather than raw provider errors, and disables automatic invocation logging and traces. These are source settings; provider-wide retention has not been verified.

Resetting the conversation clears its browser state, not information already sent to an infrastructure or AI provider. Cloudflare processing and retention depend on applicable policies and configuration; a provider-wide retention or training guarantee has not been established by this review.

Do not send passwords, tokens, private keys, license links, screenshots or document contents. Use a general description or synthetic example. AI replies can be inaccurate and are technical information, not legal advice.

Alma voice assistant (ElevenLabs)

The home pages offer Alma, an optional voice assistant (with text available on request). It is provided by ElevenLabs (ElevenLabs, Inc.), which acts as a third-party processor. Alma only starts when you open it and, for voice, when you grant microphone permission in your browser. Nothing is sent to ElevenLabs if you do not start a conversation.

When you start a conversation, your voice audio, any messages you type and the resulting transcript are sent to and processed by ElevenLabs to provide the assistant. Conversations are recorded and automatically deleted after 30 days. The widget is loaded from ElevenLabs’ published package on unpkg.com.

Do not share passwords, tokens, private keys, license links, payment details or other sensitive information with Alma. Alma cannot access your computer, your license or customer records, and its replies can be inaccurate.

Relay and external processing

Installed ALL sends requested tool data between your computer and the authorized ChatGPT session through a relay. This can include file excerpts, paths, terminal results, UI state and screenshots. ChatGPT receives the results returned to the conversation. Optional AI-backed diagnosis, repair and vision may send relevant prompts, evidence or images to Workers AI. Local execution is not a promise that all data stays on the computer.

The relay terminates TLS and sees MCP payloads in plaintext or Base64 representation. Base64 is not encryption. TLS protects individual network hops, not contents from the relay. No end-to-end encryption (E2EE) is live. A compromised relay can read, save or modify payloads and reuse observed OAuth bearers subject to local tool authority, which can include command execution.

No deliberate relay body archive was found in application source. The deployed relay and its reverse proxy keep access logging off, and synthetic probes confirmed that query strings are not logged; host crash collection, swap and Cloudflare or provider retention were not exhaustively audited, and rotated archives from before September 29, 2026 may still contain expired pairing URLs pending an operator purge decision. We therefore do not make a blanket no-logs, no-storage or zero-retention claim.

Several computers and file transfers

Activity numbers and file transfers need ALL 0.7 or later on the computers involved; earlier versions report no activity numbers and cannot send or receive transfers.

Several computers. One ALL account can have several computers. ChatGPT can list them (name, system, ALL version and when each last reported) and, for a fixed list of operations, point a request at another computer of the same account, but only where ChatGPT itself holds a live authorization on that computer. Authorization is granted separately for each computer when you connect ChatGPT to it; it is not extended to the rest of the account, and “Disconnect ChatGPT from this computer” on your activity page revokes it. A license includes up to three computers; more can be added on request (extra seats), and the limit shown on your license page is the one that applies.

File transfers. When ChatGPT moves a file from one of your computers to another (transfer.send and transfer.receive), the sending computer encrypts the file with AES-256-GCM using a key generated for that transfer only and uploads the encrypted copy, in parts of up to 16 MiB, to ALL storage on Cloudflare R2; the receiving computer downloads it, decrypts it and verifies it before saving it. Storage holds the encrypted data, its size and checksum, your license identifier and the time. It never receives the file name, the file in the clear or the key. The copy can be downloaded for at most 24 hours; after that it is deleted (deletion by the storage expiry rule can lag by up to one more day), and the unfinished parts of an upload that was interrupted are discarded by the storage provider’s own cleanup within seven days, without ever being downloadable.

The key is not kept in storage, but it does travel: ChatGPT carries it from one computer to the other inside the conversation, so it passes through the relay and ChatGPT like any other tool data (see Relay and external processing). Encryption therefore protects the stored copy; it is not end-to-end encryption against a compromised relay, and none is live. Moving a file needs ChatGPT authorized on both computers.

Activation measurement and token counter

To learn where people drop off between “I started installing” and “I use ALL every week”, ALL measures a short chain of steps. Only aggregatable data is kept: the license identifier that already exists, the time each step first happened and a source label. It never keeps file names, paths, commands, tool names or arguments, results, error messages, IP addresses or browser details.

• Steps: install started, finished or failed (with the numbered stage it stopped at, never the message), ChatGPT connected (the relay sees ChatGPT finish authorizing), first request, tenth request and active on day 7 (a call at least seven UTC days after the first). It also notes when a client other than ChatGPT finished authorizing. The step from trial to paid is read from the trial records that already exist.
• How: the installer reports its own progress with your license key; the relay reports, with the computer’s own credential, that ChatGPT finished authorizing and that the first and tenth requests arrived. The relay only classifies a request it was already forwarding: it stores and sends none of its content, only the name of the step. With the environment variable ALL_FUNNEL=0 the installer reports nothing.
• Source: when the free trial is created, the page may send the utm_source or the site that referred you (the site name only, never the path or parameters), as a short label of up to 40 characters.
• Purpose: nothing changes for you; we see counts per day and per source and fix the step where most people stop. It is not sold or used for advertising.
• Retention: each license and computer record deletes itself 180 days after its last step. Once the record of a trial that did not become a purchase is deleted (30 days after it ends), the step record that remains is only an identifier with no email or name.

Token counter (an estimate). ALL cannot see the whole ChatGPT conversation; it can only measure what passes through ALL: the arguments of ChatGPT’s calls to your computer and the results ALL returns to it. Each computer estimates them (about one token per 4 characters) and keeps them, per day, in its private folder. Only two whole numbers reach the control plane, inside the health heartbeat: those of the current day. It is not ChatGPT’s total usage. The dollar figure “equivalent to ~USD X through the API” uses the public list prices of gpt-6-sol (USD 2 per million input tokens and USD 10 per million output tokens), is only a reference and only for the tokens that passed through ALL. The server keeps each day for up to 400 days; your activity page shows the month’s total per computer.

Intentional local storage

ALL deliberately persists local tasks, workflow definitions, progress, events and results to support resuming work. Files created by tasks, job output, artifacts and configured browser profiles can also remain on the computer. This product storage is separate from relay telemetry.

ALL 0.7 and later keep the records below in ALL’s private state folder on each computer. ALL does not upload them to ALL servers; what ChatGPT reads from them through a tool call is a tool result and travels as described above.

• Journal: for each file change, commit, command, job start or end, application opened or note ChatGPT leaves, the time, the kind of event, a short summary (a file name, the start of a commit message or command line, a job title), up to five paths and the working folder. ALL itself never records file contents, command output or the computer’s environment variables; a command line is recorded as typed, with recognizable secrets masked on a best-effort basis, which is not a guarantee. A note holds whatever short text (up to 300 characters) ChatGPT chooses to write in it, masked the same way. The file is cut to its most recent entries when it passes about 1 MB.
• Recipes: commands you confirmed work, per project and task (for example install, dev, build, test, deploy or release), with their working folder. Values that look like secrets are replaced by *** and a command carrying a recognizable credential is refused, so those values are not stored. ALL saves a recipe only when ChatGPT calls the save tool, never by itself, and a recipe stays until it is forgotten.
• Project index: for projects inside the folders you authorized, the name, path, kinds of project detected, git remote without credentials, default branch, last-modified time and build commands read from files such as package.json or Makefile. It never holds source code or other file contents, and it is rebuilt from time to time.
• Job records and summaries: for each background job, the title, the command that was run, folder, status, exit code, timing, the size and checksum of the files it was expected to produce, a digest of up to 1,500 characters of its last output lines and, next to the record, the job’s full output log. Unlike the journal, these are not scrubbed of secrets: whatever a command line or its output contains is on disk. The 200 most recent job records are kept, plus any job still running, cleaned up when a new job starts.
• Operation receipts: to avoid running a repeated request twice, the private state folder keeps for 24 hours a receipt of the reply of each operation except those of the file reader (files.extract), job status (jobs.status, jobs.wait, terminal.job.status), the journal (journal.recent), code references (code.references) and git review (git.review). Those replies can contain file content and command output (reading a file, searching, running a command, a git diff). Every reply, and the copy of it that a receipt stores, is passed through one function (redact_public_result) that replaces text matching a list of known credential formats with [credential redacted]; that list is empty in this version, so nothing is replaced, and a receipt holds the reply as returned. It can therefore contain secrets that were in a file or in the output of a command. Receipts are never uploaded; what ChatGPT reads is a tool result and travels as described above.
• Extracted text and media: the transcripts, OCR text and video frames that files.extract writes for long jobs are file-derived content. They stay in a folder inside the private state folder for up to 7 days and are then deleted.

Removing a device from a license does not erase those files, task records, browser data, backups or copies already sent to ChatGPT or another provider. There is no universal local deletion deadline in this source review. Review the relevant local storage and provider controls before assuming data has been erased.

Payments, Stripe and billing

Stripe processes payment data. ALL does not receive or store your full payment-card number. We receive information needed to confirm payment, identify the purchase or subscription, issue the license, enforce the plan and respond to states such as active, past_due, canceled or similar where applicable.

We may retain Stripe identifiers and commercial records while needed for the contractual relationship, support, fraud prevention, accounting, tax or legal claims.

Analytics, cookies and similar technologies

The site uses Cloudflare infrastructure and may use Cloudflare Web Analytics for aggregate performance and referral metrics. Checkout uses a secure, HttpOnly, same-site browser cookie to bind the returning browser to its purchase. Chat history is not stored in that cookie.

We do not use the site to sell personal data or build cross-site behavioral advertising profiles. Stripe and other providers may use their own technologies on their domains; their practices are independent. This source review does not verify every deployed analytics setting.

Retention and unknowns

Retention depends on the data category and purpose. License, device, purchase and subscription records may remain during the service relationship and afterward for support, fraud prevention, accounting, tax or claims. Local tasks and results intentionally persist; they are not automatically erased when a short authentication session expires.

Activity numbers: the latest device health record is kept up to 30 days and each day’s operation count up to 400 days. An encrypted transfer copy can be downloaded for at most 24 hours. On the computer, the journal keeps its most recent entries (cut at about 1 MB), job records keep the 200 most recent job records plus any job still running (cleaned up when a new job starts), and recipes and the project index stay until removed or rebuilt; none of these local files is erased when a device is removed from a license.

Some application counters and event-deduplication records have technical expiry settings. Those settings are not a complete retention schedule and do not determine retention in logs, backups, proxies or AI providers. Relay access logging is disabled and verified, but host crash collection, swap and Cloudflare or provider retention were not exhaustively audited, and rotated relay archives from before September 29, 2026 may still contain expired pairing URLs pending a purge decision. This page does not invent a provider-specific deletion deadline or promise zero retention.

For a privacy request, use the contact below. Applicable legal obligations and exceptions may affect what can be deleted.

Free trial: the keyed hashes of the email and of the computer are kept for 24 months, so the trial is one per person and per computer; an unsubscribe from trial offers is kept for the same 24 months. 6-digit codes live 10 minutes. The record of a trial that did not become a purchase (it holds the email, the language and the marketing choice) is deleted 30 days after the trial ends or, if it never started, 30 days after its 14-day activation window closes.

Activation measurement: each license’s and computer’s step record deletes itself 180 days after its last step; each day’s token numbers are kept up to 400 days.

Remarketing and communications

If you voluntarily submit the ALL updates form and check the consent box, we may use your name, email, use case and visit-attribution data to send relevant ALL capability updates, launches and offers.

The marketing box is not pre-checked, and purchasing ALL does not automatically subscribe you to promotional communications. We record the consent version and timestamp so we can document when consent was given. You may withdraw it at any time by emailing admin@fynory.app; withdrawal does not affect processing that was lawful before withdrawal.

We may also retain attribution parameters such as UTM values, referrer, language and landing page to understand which campaigns generate interest. We do not sell this information to advertisers.

When you start a free trial, the offers box is also optional and unticked. The trial’s service emails (welcome, 6-digit code, reminder and end) are always sent, because they are necessary to provide the trial and are not marketing. Later offers (at most two) are sent only if you ticked the box, carry a one-click unsubscribe link and stop as soon as you choose a plan.

Service providers and disclosures

We may share information with providers that process data for us or are necessary to deliver the service, including Cloudflare for infrastructure/AI, Stripe for payment and ElevenLabs for the optional Alma voice assistant. We may also disclose information when reasonably necessary to comply with law, protect rights, investigate abuse or fraud, complete a corporate transaction, or protect users and systems.

We do not sell personal information to advertisers.

International transfers

ALL is operated by a United States company and uses global infrastructure providers. Information may be processed in countries other than your residence. Where required by law, we seek to rely on recognized transfer mechanisms or contractual safeguards available through our providers.

Security and deployment state

The September 29 hardening ships in tiers. Control-plane and relay protections are deployed and production verified: five-minute Ed25519 device sessions with proof of possession and in-band renewal, control-only signing with public relay verification, per-device proof, replay, generation and revocation checks, and a hardened relay with no HMAC or master secret and access logging disabled. Device-side protections — the persistent local identity key, automatic short-session renewal and hardened OAuth/PKCE with hashed secret state — are available in runtime 0.6.7; active after updating; they activate when a client installs the updated runtime, with automatic migration and no periodic reinstall or re-pairing. End-to-end encryption is not implemented.

Those changes address authentication, not end-to-end content encryption or erasure. No system is guaranteed invulnerable. See Security for the trust boundary, the tiered controls and remaining risks.

Your privacy rights

Depending on where you live, you may have rights to access, correct, delete, restrict, object, port or appeal certain processing. You may also withdraw consent where consent is the applicable legal basis.

Start a request at admin@fynory.app with a non-sensitive description. Do not send credentials, private license links, screenshots or document contents. Identity verification may be needed to protect the account; this public chat is not an identity-verification channel. Applicable rights, deadlines and exceptions depend on the law.

Children

ALL is not directed to people under 18 and is not designed for a minor to purchase a license independently. If you believe we received information from a child improperly, contact us.

Third-party services and links

ALL can work with third-party applications and accounts you choose to use. Those providers control their own data, authentication, cookies, terms and policies. Authorizing ALL to operate an existing session does not make Fynory LLC responsible for that provider’s independent practices.

Changes to this policy

We may update this Policy when ALL, the law or our providers change. We will publish the revised version and its updated date. If a change requires additional notice or consent by law, we will use an appropriate mechanism.