Privacy policy
This is a translation for convenience. If it differs from the Chinese version, the Chinese version governs.
This policy explains what personal information the still.fail hosted service collects, how it's used, who it's shared with, how long it's kept, and what rights you have. Please read it carefully, especially the parts in bold.
1. Who processes your information
still.fail is run by an individual developer based in mainland China (“we”). For any questions or requests about personal information, email hi@still.fail.
2. Which services this policy covers
This policy applies to the still.fail hosted service we operate, including the still.fail website, the web app (app.still.fail), public share pages (share.still.fail), preview pages (preview.still.fail), and the desktop, Android and iOS apps when they connect to these services (together, the “Service”). It also applies to the beta channel youdid.wtf.
This policy does not apply to:
- Stations on your own machines: a station runs on a computer or server belonging to you (or to a member of your workspace). Chat history, files and the agents' work are stored there, under the control of you and your workspace, and we can't access them. So that losing one machine does not lose them, each station also keeps a copy of its chats (their messages, the files attached to them, and their sessions' records; not the agents' transcripts, working folders or signed-in accounts) on two other stations of the same workspace, and never on another workspace's stations. A new station of the workspace can take a lost one's chats back from these copies. The copies go between your stations end to end encrypted, never through our servers;
- Services that others deploy themselves from the still.fail open-source code;
- Third-party services you use on your stations, such as Claude, ChatGPT/Codex, Hermes Agent and Pi and the model services you configure for them, Google's Antigravity CLI, Slack and GitHub. Their own privacy policies apply.
3. The most important point first: your chat content isn't stored with us
still.fail's agents run on your own stations. Your devices connect directly to your stations with end-to-end encryption; when a direct connection isn't possible, traffic is forwarded through a relay server, which only passes on encrypted data and can't see the content. Our servers don't store your chat content, attachments or files.
There are two exceptions, where content passes through or is stored with us:
- Push notifications: to show notifications on your devices, your station sends the chat title, the sender's name and the first 140 characters of the message through our servers to Google's Firebase Cloud Messaging (Android), Apple Push Notification service (iOS) or your browser's push service (web). We only forward this text; we don't store it. On a device where you turn off “Show content in notifications”, its notifications only say what happened (such as “New message”), and the message's text is not forwarded;
- Issue reports: issue reports that you or the agents on your station choose to submit (title, description, context, the affected desktop OS when known, and logs) are stored on our servers so we can investigate. The still.fail team's replies on a report, and your agent's replies to them (their text, time, who replied, and whether a team reply reached the conversation the report came from), are stored with the report; the team's replies go to that conversation through your station.
Publicly shared chats are also stored on the publisher's station; visitors' browsers read them directly from the station over an encrypted connection. The key after the # in a share link is never sent to our servers.
When you use shared debugging, the station keeps the time the chat last actually used debugging while its grant is active and sends it to your devices over the encrypted connection, to display and sort the list. When the grant ends, that time is removed from the grant record. Recording that time does not store the contents of debugging commands.
A station can be set to tasks only: it runs no agents and keeps no chats, only runs tasks for the workspace's other stations, and belongs to the member who added it or set it so. Such a station keeps on its own machine the grants its owner signed for it (the chat's title, the email of who started it, the kind of grant and until when) and a record of the commands it was told to run (the time, which station's chat asked, and the first 2,000 characters of the command). A grant is removed from the machine when it ends, and only the number of one taken back is kept until it would have expired; once the record passes 4 MB only its two most recent files are kept. The station of the chat keeps the request and its outcome. When a full station is set to tasks only, the chats it already holds stay on the machine and are not served until it is a full station again. All of this travels over the encrypted connection between your devices and the stations, and never through our servers.
Members can attach devices (phones, boards, cameras, routers, serial ports and the like) to a station for agents to use. The station a device is attached to keeps its name, kind, what it is reached at (an address and port, a serial port's path or an adb serial), the member who added it and when; one attached to a tasks-only station also keeps the grants and the no-approval setting its owner signed for it, and each connection to it goes into that station's command record. When a chat connects to a device, the data travels over the encrypted connection directly between the chat's station and the device's station, never through our servers, and is not kept. When a device is removed, this is deleted from the station.
Your presets and the models you starred are kept on every station of the workspace that your devices reach, so they are the same on all your devices: each preset (its name, folders, instructions for the agents, the model and stations its chats are opened with, who made it and when, and how many times it was changed) and, of a deleted one, only its identifier, whose it is, how many times it was changed and when it was deleted (the latest 500 in each workspace), so that an old copy elsewhere does not bring it back; of the starred models, each model's identifier, whether it is starred and when. Your devices carry them to the stations over the encrypted connection, never through our servers. The workspace's members and agents can see the presets a station keeps.
When you use Add › Slack account in your vault, you allow it on Slack's page and Slack sends your browser back to a page of the web app (/vault/oauth). That page passes the code, inside your browser, to the device that started the sign-in. Loading the page goes through our static web server with the code in its address; the server keeps none of it, and the code is of no use without a check value that only the starting device holds. That device gets your own token from Slack directly and asks Slack which Slack account it belongs to, then keeps the token, your Slack user name and your Slack workspace's name in your vault, the same way as a token you add by hand. While a sign-in is under way the device keeps a one-time check value for an hour at most. Removing the app's access in Slack, or deleting the item from your vault, ends the token.
When a Comma Router writes to a chat, the station keeps the idempotency key and the chat and message numbers on its own machine so a retry does not write twice. Until delivery is handed on, it also temporarily keeps the sessions that were to receive it. Records older than 31 days are removed in batches on later writes; while idle, they remain until later writes. The deduplication record does not copy the message text.
When connected to Comma, the station keeps an index of usage and chats, including chat titles, whether a Comma Router has written in each chat and, for a chat that waits for someone (a person's answer, a finished result its starter has not read, a failure), the email addresses of who it waits for, the agent's one sentence on what is needed, done or failed, and the chat's link in still.fail, on its own machine for each workspace a Comma connect is of (its home workspace or one it joined), to send to Comma. Refreshing it updates statistics from the last 31 days; older index rows and index files left by earlier versions may remain until the station's data is deleted. Changing workspaces does not automatically delete them.
To draw how an allowance changes and how fast it is spent on Profile's overview, a station keeps on its own machine, for each account, what it read of the subscription's allowance each hour (each window's name, the percent used, when it resets and how long it is). It keeps at most the last 192 hours (8 days) of each account, removing older records as a new hour is written, and removes them when the account is deleted. The account's calls each hour by model (how many calls and tokens) are not kept again: they are read for the last 192 hours from the usage statistics by quarter hour the station already keeps. They go only over encrypted connections to the devices of your workspace's members, never through our servers, and hold nothing of the conversations.
When you create a Feishu bot by scanning, the station temporarily stores the scan link, device code, creator, chosen connection settings and bot credentials returned by Feishu on its own machine, so it can resume saving after a restart. Once the connection is saved, the temporary credentials and link are removed; completed or failed registration records are deleted 24 hours after the scan expires. If the bot is created but its connection has not been saved, the credentials are kept for retry. The connection's own credentials are stored in the station configuration.
4. What we collect
- Account information: if you sign in with Google, your Google account ID, email address, name and avatar URL; if you sign in with an email code, your email address. We also keep when your account was created and last active, and its plan, eligibility and ban status;
- Sign-in sessions and devices: device or client names (such as a computer's name, or a phone's as set in its settings), when sessions were created and when they expire, and when devices were first and last online;
- Workspace information: workspace names, members and roles, invitations (including invitees' email addresses), invite codes, and the names, icons, versions, online times and mode (full or tasks only) of stations that have joined, with the owner of each tasks-only station;
- Push tokens: the tokens or push endpoints your devices or browsers register to receive notifications, with that device's notification settings (its language, which kinds of notifications it shows, whether it shows their content);
- Issue reports: see section 3;
- Network and log information: the IP address, request path, time and browser information (User-Agent) when you access the Service, recorded in Cloudflare's request logs; relay servers can see connections' IP addresses, node public keys and traffic volume. To prevent abuse, we rate-limit sign-ins using salted hashes of IP addresses and email addresses;
- Usage statistics and error information: the web app and the Android app send PostHog action events, counts, durations, program errors and stack traces, and the web app also records session replays (all text, input and images on the page are masked at recording time). The program errors the iOS app sends are the crash and hang reports iOS gives it the next time it starts (Apple's MetricKit: stack addresses, the kind of exception, the app version, the device model and the system version), kept on the device until you sign in, then sent; the Mac app too sends the program errors it met and got past. The apps on every platform (Android, iOS, macOS, Windows, Linux and the web) also send PostHog, every so often, how many times features were used: which features (only the action names written in the program, such as sending a message, archiving a chat or changing a setting), which kinds of page were opened (only the kind, such as a chat or which page of the settings, never which chat, station or file), how many times each, and the app's platform and version; never what an action carried or its parameters, what was typed or any message's content. The Android, iOS and Mac apps also send records of the interface stalling: which kind of work held it (input, drawing, a system callback), how many milliseconds each took, how many elements were drawn, made and laid out and how many pieces of text were measured, and how many stalls there were while the app was in front and how many of its frames took how long, and which kind of page was showing when it stalled (a chat, the list of chats, which page of the settings), those counts split by it too, and whether the interface was laid out for a small, medium or large screen and the window's width and height when it stalled: only the kind of page, never which chat, station or file, and without what was typed or any message's content. These two apps also send how large the data the app received inside itself was (what a page received as the Android app switched to it, and what came and was never used): only the names of the kinds of data and of their fields, byte counts and message counts, never the data itself. This data is tied to your account ID only, without your email address or name. You can turn off “Send usage and error reports” in the app's settings; once it's off, that device sends nothing to PostHog. Clients, stations and servers also send request traces to Google Cloud, including account ID, session ID, request path and duration, and the way each connection goes, its round trip, and how long the app was off screen before it came back. These statistics and traces don't include chat text, file names, prompts, agent output or conversation titles. Traces are used to diagnose failures and performance problems and aren't affected by that switch. Stations also send their run logs to Google Cloud by default (what the station did and when: its connections to still.fail and to other stations, errors, versions), with home directory paths, email addresses, things that look like keys or tokens, quoted text and fields that may hold what someone wrote removed first; they don't include chat text or file contents. The workspace's admins can turn this off with “Send run logs to still.fail” under Diagnostics in the station's settings in the app, or in the station's
config.json("telemetry": { "logs": false }). Station error reports to PostHog are sent only if the station's user turns them on; they're off by default.
The still.fail website (still.fail) itself doesn't use analytics.
5. How we use it
- To sign you in, identify your account, devices and workspaces, and let your devices find and connect to your stations;
- To send sign-in codes and notifications, and the email telling someone added to a workspace who added them and to which workspace;
- To enforce plan limits, prevent abuse and attacks, and keep accounts secure;
- To troubleshoot problems, understand how features are used, and improve the Service;
- To respond to your questions and requests, and to meet legal obligations.
We don't sell your personal information, use it for advertising, or use your data to train AI models.
6. Cookies and local storage
- Signing in uses a cookie valid for 10 minutes, to complete sign-in securely;
- The web app stores sign-in credentials and app data in your browser's IndexedDB, and interface preferences (theme, sidebar width and so on) in localStorage;
- PostHog analytics in the web app stores its own identifier in your browser, and whether you turned statistics off.
We don't use third-party advertising cookies.
7. Who we share it with
To provide the Service, we engage the following service providers to process some personal information. They process it only as we instruct, for the purposes listed in the table below:
| Provider | Location | Purpose | Information involved |
|---|---|---|---|
| Cloudflare, Inc. | United States; global network | Hosting the website, servers and data storage; request logs; sending verification code emails and emails about being added to a workspace; relay servers | All information stored on our servers listed in section 4, request logs, connection metadata |
| Google LLC | United States | Google sign-in; Android push (Firebase Cloud Messaging) | Google account information, push tokens and notification text |
| Apple Inc. | United States | iOS push (Apple Push Notification service) | Push tokens and notification text |
| Your browser's push service (e.g. Google, Mozilla, Apple) | Varies by browser | Web push notifications | Push endpoints and encrypted notification content |
| PostHog, Inc. | United States | Usage statistics, error tracking, session replay | Account ID, usage events, error information, masked session replays |
| Google LLC (Google Cloud) | United States | Request tracing, station run logs | Account ID, session ID, request paths and durations; station ID and its scrubbed run logs |
| Alibaba Cloud | Beijing, China; Hong Kong, China | Relay servers, to improve connections in mainland China | Connection metadata (IP addresses, node public keys, traffic volume), no content |
Beyond this, we disclose your information only when required by laws and regulations, when lawfully requested by law enforcement or judicial authorities, or when genuinely necessary to protect the life or property of you, other users or the public. Other members of a workspace can see your name, email address and avatar.
8. Cross-border transfer of personal information
Because Cloudflare, Google and PostHog in the table above are located outside mainland China, your account information, device and workspace information, push notification text, usage statistics, error information and logs are transferred outside mainland China (mainly to the United States) to be stored and processed for the purposes listed in section 5. You can exercise your rights against us or these recipients as described in section 10. Contact details for the overseas recipients are on their websites. By signing in to and using the Service, you consent to these cross-border transfers; if you don't agree, don't use the Service.
9. How long we keep it
- Sign-in codes: 10 minutes, deleted once used;
- Sign-in sessions: up to 30 days; they expire after 7 days without use;
- Invitations: 7 days;
- Whom an email about being added to a workspace went to (the workspace, the email address and the time): used for a 24-hour sending limit. Expired records are cleared the next time members are added, and records are also cleared when their workspace is deleted. If sending fails, we separately keep the workspace and email address as pending retry state, so adding the recipient again resends it; this record contains neither the mail text nor the reason for failure. This state is removed when a retry succeeds, the membership or pending membership is removed, the workspace is deleted, or the recipient deletes their account. Otherwise it is kept only while that membership or pending membership exists;
- Push tokens: deleted when you sign out or the device stops receiving notifications;
- Account, workspace and station information: kept while you use the Service, and deleted as soon as you delete your account or the workspace;
- Issue reports and their replies: kept as needed after the issue is resolved, deleted along with your account, and you can ask us to delete them at any time;
- Request logs, usage statistics, traces and station run logs: kept for the retention periods of Cloudflare, PostHog and Google Cloud respectively, then deleted automatically.
Where laws or regulations require a longer retention period, that period applies.
10. Your rights
You have the right to access, copy, correct, supplement and delete your personal information, to ask us to explain our rules for processing personal information, and to withdraw consent and delete your account.
- Workspace owners can delete a workspace in the app;
- Deleting your account: choose “Delete account” in the app's account settings. Your account information, sign-in sessions, push tokens and issue reports are deleted at once; workspaces with only you in them are deleted too; you're removed from the other workspaces you're in. If you're the only owner of a workspace that has other members, first make someone else an owner or delete that workspace. Deletion can't be undone;
- Turning off usage statistics: turn off “Send usage and error reports” in the app's settings;
- For other requests, also email hi@still.fail. We'll reply within 15 working days.
Your chat history and files are on your stations, where you can view, export and delete them directly. A chat you delete is deleted from its copies on the workspace's other stations as well, the next time each is brought up to date; a station that leaves a workspace deletes the copies it kept of that workspace's stations.
11. Security
Communication between devices and stations is end-to-end encrypted, the website uses HTTPS throughout, only hashes of sign-in credentials are stored, and access to servers is tightly restricted. But no system is perfectly secure; if a personal information security incident occurs, we'll notify you and take remedial measures as required by law. If you find a security vulnerability, please report it privately as described in SECURITY.md.
Agents running on a station have the permissions of the corresponding system account on that machine. Run stations on machines and accounts you trust.
12. Minors
The Service is for users aged 16 and over. If you're under 16, please don't use the Service. If we find that we've collected personal information from a child under 14 without a guardian's consent, we'll delete it as soon as possible.
13. Changes to this policy
We may update this policy. Updates will be posted on this page with a revised effective date. We'll tell you in advance, in the app or by email, about material changes (such as adding a new third-party service provider or changing the purposes of use).
14. Contact
Email: hi@still.fail