Ixel
Your data

Privacy

Ixel has no account, no server of its own, and nothing that reports back to us. It runs on your computer and talks only to the services you set up, when you use them. This page lists every one, and everything it keeps.

In short

  • No telemetry. Ixel sends no analytics, crash reports or usage pings. We don't know who uses Ixel or how.
  • No Ixel account, and no Ixel server. Your questions go from your computer to the AI companies you chose, and nowhere else.
  • Your keys stay yours. Each key goes only to its own company, and never over plain http:// except to your own computer (and, for a git host's token, across your Tailscale network, which encrypts it). On your computer they're encrypted, and the key that opens them is in your system's keychain. On a computer with no keychain Ixel can use (a Linux server, say), they're in a plain-text file only you can read, and Health says so.
  • Private keeps it at home. Turn on Private in Settings, and only the models on your own computers answer; every other model sits out. Your questions and their answers stay on your computer or your network, as long as your servers run the models themselves.
  • Some companies train on what you send. It depends on the company and how you connect, usually an API key or a sign-in plan. See which ones, and how to keep yours out.
  • Little is kept, and not for long. Ixel keeps your last ixel review questions for a day, a log of what Machines did for 30 days (never a command's text), and none of the app's conversations once it stops. ixel forget clears it all. See the full list.
The companies you use

What you send to Anthropic, OpenAI, Google, xAI or any other model company is covered by that company's terms, not ours. Ixel can't make them forget it. Which companies may train on it says what each one does. For questions that must not leave your computer, use a local model and turn on Private.

What goes where

Everything Ixel and Handoff send over the internet, and when:

ToWhenWhat's sent
The companies of the models on your panel, through your API keysWhen you ask a questionYour question, up to three earlier questions and answers in the same conversation, code or files you attach, and in later rounds the other models' answers. Pictures you attach go only to the models that can see them.
Claude Code, Codex, Gemini CLI, Copilot, OpenCode, Grok Build, or Antigravity once you switch it onWhen you ask a questionThe same, handed to the program on your computer, which sends it to its company. Claude Code and Codex are told not to save it; for Gemini CLI, Copilot and OpenCode, Ixel removes what each run saved once it's done, and Grok Build saves it only in a folder Ixel gives each run and then deletes (what each keeps). Ixel turns off OpenCode's share links and Copilot's copy of the session to GitHub's cloud. Whenever Ixel runs OpenCode, it also stops OpenCode from fetching its model catalog from models.opencode.ai.
Google, from Gemini CLIWhen Gemini CLI answersUsage statistics: how long the question was, the model, which tools it called, your system and its version, with your Google account's email or a random id, but never the question. Gemini CLI has no switch for one run, so only you can turn them off, with "privacy": {"usageStatisticsEnabled": false} in ~/.gemini/settings.json.
GitHub, from CopilotWhen Copilot answersCopilot's help says it sends telemetry, and its only switch for that also turns off its GitHub sign-in, so Ixel can't turn the telemetry off. We haven't checked what it sends.
xAI, from Grok BuildWhen Grok Build answersNothing besides the question: Ixel runs it with its usage statistics, traces, feedback and updates off. It does ask the model, in a short call of its own, for a title for the session, which counts against your plan; the title is deleted with the session.
The company of the model a Handoff worker runs on: Claude Code, Codex, OpenCode, Grok Build, Hermes Agent, Kimi Code, Cline, OpenClaw or an agent you addedWhen a task you approved runsThe task as you approved it, with its history, and whatever the agent reads in that task's copy of your project. An agent that runs with its own permissions (Kimi Code, Cline, OpenClaw) can reach what you can, so it runs only once you switch it on. A model on your computer sends nothing out.
Anthropic, through your Claude Code, when Ollama on this computer doesn't have qwen3:4b-instructWhen Auto crew picks a crew (handoff crew, handoff approve --auto, or Auto on the Board)The task's title and text, up to 4,000 characters, to Claude Haiku 5.5 with no tools and from an empty folder, so it reads none of your project. With that Ollama model, nothing leaves your computer.
The company of the model you pick on the Agent pageWhen the Agent works on a goalYour goal and follow-ups, its memory files and the skills it loads, what it reads in the folder you picked (or, with Computer access on, your home folder), the output of commands you approve (which can include files outside that folder; on a Mac or Linux, not your secret folders, keychains or browser data), and pictures you attach or it looks at. A model on your computer sends nothing out.
GitHubWhen you ask the Agent to add a tool from a GitHub link, or turn on an add-onA download of that project at the version its card names. Nothing about you; GitHub sees your IP address.
The company of the model you pick for a diagramWhen you ask with Diagram onYour question and the Diagram Design add-on's instructions and templates. A model on your computer sends nothing out.
A model server on your computer or networkWhen you ask a questionThe same, kept between your machines. No key, unless you gave that server one in the settings file.
Model servers on this computer, or at the address you typeWhen you press Look in Settings or run ixel setup, and each time Settings opens, for the models you addedA request for the list of models it has. Nothing of yours, and no key unless you gave that server one. Look and setup ask it directly, never through a proxy, and never search your network.
ollama.com, or Hugging Face for an hf.co/ name, from your OllamaWhen you press Get under Get a modelYour Ollama downloads the model, as ollama pull does; they see that computer's address and the model's name. Ixel itself talks only to your Ollama, and sends nothing else.
The companies your API keys are forWhen Ixel picks the newest model (latest), lists models in Settings or setup, or checks your keysYour key, to get the list of models. No question.
TypeSafe AIOnly if you choose TypeSafe for TriageYour question, up to three earlier questions, and the models' answers. Never your other keys, or which model wrote what.
OpenAI or GroqWhen you use Sound, with Private offThe recording or sound file, to be written out. Ixel doesn't keep it.
xAI or OpenAIWhen you make picturesYour description. Files you name are read by one of your chat models first, and only its description goes on.
Your project's git host (GitHub, GitLab, Codeberg, or your own)When you open a project's Board, or press Review… or Fix… on a pull requestThe repository's name, to list its pull requests, and your token for it if you saved one. Review… and Fix… fetch the pull request with git, using your own git login.
GitHubWhen the terminal chat starts, at most once a dayA check for a new version, with git. No Ixel data; GitHub sees your IP address. Turn it off with [updates] check = false.
GitHubWhen you run handoff rankings updateA download of Handoff's public model rankings. Nothing about you; GitHub sees your IP address.
ixelai.com, GitHub and the Python Package IndexWhen you install or run ixel update or handoff updateDownloads, like any install.
Your own serversWhen you press Check key, Connect or Run on machinesAn ssh connection you started.
Google, from the app window's browserWhen the app window opens, on Windows and LinuxIxel turns off the browser's background services, which stops its update, autofill, time and search requests. Two are left that no switch stops: its push-message service checks in, with the browser's version, system and language, and its sign-in service asks which Google accounts the window is signed in to (none, unless you signed in to Google there). Neither has anything from Ixel's page. If your computer uses a public DNS server with a secure version, such as 8.8.8.8, the browser also looks names up there over HTTPS. We measured this in a Chromium window; Edge, the window on Windows, has Microsoft services of its own that we haven't measured.

Handoff's plugin talks to your AI apps on your computer only. Handoff itself uses the network for handoff update and the rankings update listed above. It also sends requests to the model server a local worker uses, which can be on another computer and gets a key only if you set one. Otherwise, it reaches the internet only through the programs it starts for you (an agent run, or Ixel).

The app's page and its server run only on your computer. The server listens on 127.0.0.1 with a one-time key, and the page can't load anything from the internet: no fonts, no scripts, no images from elsewhere.

Which companies may train on it

Once your question reaches a company, Ixel can't keep it private. That company's rules decide whether it keeps what you send, uses it to train its models, or lets its staff read it. Mostly it comes down to how you connect: API keys on a paid account usually aren't trained on, while sign-in plans and free tiers often are.

Our part, and theirs

Ixel sends a question only where you tell it to and keeps no copy on any server. What a company does after that is set by its terms, which you agreed to with that company. We can't change them, so we tell you what they say, and the choice is yours.

This is what each company says on its own pages, as we read them on October 4, 2026. They change often, so check the linked page before you rely on it. Where the pages we read don't answer, the table says "Not stated", and where we didn't look, "Not checked".

Where it goesUsed to train models?Read by people there?Kept for
Anthropic, with an API keyTraining (updated Aug 18, 2026), how long (Jul 1, 2026)No, unless you send it feedback or agree to share your data.Not stated on its API pages.Up to 30 days. Up to 2 years if flagged under its Usage Policy, and its safety scores up to 7 years.
Claude Code, signed in with Claude Pro or MaxConsumer terms (Aug 28, 2025, changed Sep 10, 2026), training and who can see it (Mar 16, 2026), how long (Jul 1, 2026)Your choice. Only if you've allowed it in Claude's Privacy settings, where you can change it any time.Not by default. Only when you send feedback, or when its Trust & Safety team reviews a flagged conversation.Up to 5 years, with your name taken off, if training is on; 30 days if it's off. Up to 2 years if flagged, and its safety scores up to 7 years.
OpenAI, with an API key (also Pictures and Sound)Enterprise privacy (Jan 8, 2026)No, by default.Authorized staff, for support, abuse and legal reasons, and outside contractors who review for abuse.Up to 30 days, to run the service and catch abuse. Some uses can ask OpenAI to keep nothing.
Codex, signed in with ChatGPT Free, Plus or ProTraining and data controls (late Sep 2026), who can see it (mid Sep 2026)Yes, unless you turn it off. OpenAI may use it to train its models. Improve the model for everyone, in ChatGPT's Settings, Data controls, also covers Codex. A separate Codex setting, Include environments, decides whether OpenAI may also train on the whole coding environment. Business, Enterprise and Edu plans aren't trained on by default.A limited number of staff and service providers, for abuse, support and legal matters, and to improve models while training is on.Not stated for Codex. Deleted ChatGPT chats go within 30 days, unless already de-identified or kept for security or legal reasons.
Google Gemini, with a free API key (Ixel's Google key or Gemini CLI's, with no billing on its Cloud project)Gemini API terms (Apr 28, 2026)Yes, outside the EEA, Switzerland and the UK. Google uses it to improve its products and machine learning, and asks you not to send anything sensitive, confidential or personal. In those three, the paid key's rules (next row) apply to a free key too.Yes, outside the EEA, Switzerland and the UK. Human reviewers may read and annotate it, after Google separates it from your account.Not stated.
Google Gemini, with a paid API key (billing turned on for its Cloud project)Gemini API terms (Apr 28, 2026)No.Not stated. Google logs it to catch abuse.For a limited period of time; no number given.
xAI (Grok), with an API key (also Pictures)Enterprise FAQ (Feb 25, 2025)No, unless you took free credits in return for letting xAI train on your data.A limited number of authorized staff, when legally required, such as for security incidents or misuse.Deleted within 30 days, unless flagged or agreed otherwise.
Grok Build, signed in with your xAI account (SuperGrok)Grok Build docs and its own settings (version 1.0.46, Oct 2026), Grok FAQ (no date), terms (Sep 11, 2026)Not unless you opt in. Grok Build's own setting, Coding data, retention, and training (type /privacy in grok), is off by default; opting in lets xAI keep your prompts, traces and metrics for training and debugging. Separately, xAI's FAQ says it may train on what you send Grok's apps unless you turn off Improve the model in Grok's Settings, Data Controls; it doesn't say whether that covers Grok Build, so turn it off too if you want to be sure.A limited number of authorized staff, to improve its models, look into security incidents and misuse, and meet legal obligations (the FAQ, for Grok).Not stated for Grok Build. Deleted Grok conversations go within 30 days, unless taken off your account first or kept for safety, security or legal reasons.
GitHub Copilot, on Free, Pro, Pro+ or MaxGitHub's announcement (Mar 25, 2026), the settingYes, unless you turn it off (since April 24, 2026): your prompts, its answers, and the code around them. Turn off Allow GitHub to use my data for AI model training in your Copilot settings on github.com; an earlier opt-out still counts. It may be shared with GitHub's own family of companies, Microsoft included, but not with outside AI model companies. Business and Enterprise aren't trained on.Not stated.Not stated.
OpenCodeOpenCode Zen (Oct 3, 2026)OpenCode passes your question to the company you signed in to in OpenCode, so look up that company's row (we haven't checked OpenCode's own terms). On OpenCode Zen, OpenCode's own service: no for paid models, but its free models may use what you send while they're free. Ixel can't use the free ones anyway.Not stated.Zen's providers keep nothing, except 30 days for OpenAI's and Anthropic's models.
TypeSafe AI (Triage, only if you choose it)Privacy policy (Nov 19, 2025)No. It says it won't train or fine-tune models on what you send.Not stated.As long as reasonably necessary; no number given.
Groq (Sound)Your data in GroqCloud (no date)Not stated: its data page doesn't say either way.Not stated.Nothing, by default. Up to 30 days when kept to fix errors or look into abuse. Turning on Zero Data Retention, in Groq's Data Controls, stops the copies kept to fix errors.

Gemini CLI with a Google sign-in: since June 18, 2026, Google's sign-in no longer works in Gemini CLI for personal accounts (Google's notice, updated Sep 2, 2026). Use a Gemini API key (the rows above), or a work account with Gemini Code Assist Standard or Enterprise, for which Google says what you send isn't collected or used for training.

Handoff works the same way. An agent it runs, or an app you connect to the board (Claude Desktop, Grok Build and the rest), sends what it's given to its own company, under that company's rules for how you connect (a plan or an API key). Answers and reviews it asks Ixel for go to the models on your panel.

Models you add yourself, or through OpenCode:

Where it goesUsed to train models?Read by people there?Kept for
DeepSeekPrivacy policy (Feb 10, 2026)Yes. To opt out, email privacy@deepseek.com. Its privacy policy names no exception for the API.Not stated.Not checked. It's stored in China.
Kimi (Moonshot AI)API terms (Jul 30, 2026), API privacy (Apr 30, 2025), app terms (Aug 13, 2026)Yes, on every plan, the paid API included. Opting out means contacting Kimi, and it covers only what you send later; for the API it takes a written agreement.Specific employees, when needed for business purposes.As long as necessary. The API's data is stored in Singapore; the app's may be in China or Singapore, depending on your account.

We haven't checked the others you can add by hand or through OpenCode, such as Z AI (GLM), Xiaomi (MiMo), Mistral or OpenRouter. Read a company's terms before you send it anything you'd mind it keeping.

Two tools Ixel uses only if you switch them on, so read these first:

  • Google Antigravity. Google keeps your prompts, code and replies to improve its products and models, and its employees and contractors may read them. You can opt out in Antigravity's Settings, but its terms don't say whether that also stops people reading them. Its FAQ says reaching Antigravity through other tools breaks its terms and can get your account suspended, and points those tools to a Gemini API key. So Ixel leaves Antigravity off until you switch it on, and shows this warning first; a Gemini API key avoids the question.
  • Cline, which Handoff can run as a worker once you switch it on. Its free models may use what you send to improve the model, and most come from labs it doesn't name. On Cline's own provider, Cline sees your prompts and code and keeps them for a time its privacy policy (Sep 24, 2025) doesn't state. Your own API key or a local model avoids both.

Keep your questions out of training

  1. Use an API key on a paid account

    Anthropic, OpenAI and xAI don't train on API use by default, and Google doesn't once billing is on for the key's Cloud project. What you send with a free Gemini key is trained on and may be read by reviewers, except in the EEA, Switzerland and the UK.

  2. On a plan, turn training off

    For Claude Code, the training choice in Claude's Privacy settings. For Codex, Improve the model for everyone in ChatGPT's Data controls, and Include environments in Codex's settings. For Copilot, Allow GitHub to use my data for AI model training in your Copilot settings. For Grok Build, leave Coding data, retention, and training (/privacy in grok) off, and turn off Improve the model in Grok's Data Controls.

  3. Skip the ones that train with no switch to stop it

    These include Kimi and DeepSeek, unless you've asked them to stop by email or support. Free models in OpenCode Zen and Cline may use what you send, too.

  4. Keep it on your computer

    For anything that mustn't leave it, use a local model and turn on Private. Then only the models on your own computers answer; Honest limits says what Private doesn't cover.

Not training on something isn't the same as not keeping it: several of these companies keep what you send for up to 30 days to catch abuse, and longer if it's flagged.

What's kept on your computer

Ixel's files are in ~/.config/ixel-mat (on Windows, %USERPROFILE%\.config\ixel-mat). On a Mac or Linux, that folder is readable only by you.

WhatWhat's in itHow long
config.toml, config.toml.bakYour settings: models, addresses, the names of your keysUntil you change them
agent/The Agent's memory files (SOUL.md, USER.md, MEMORY.md) and skills, each conversation's goal, steps, approvals and usage, files you attached, GitHub projects it fetched to write an add-on card, and the commands you always allowUntil you delete them
keys.encYour API keys and tokens, encrypted. The key that opens it is one item, named Ixel, in your system's keychain: Keychain on a Mac, Credential Manager on Windows, your keyring on Linux. When Ixel runs where it can't ask the keychain (over SSH, or from cron), it can't open the file, so it uses none of these keys there: set the key in that session's environment insteadUntil you remove them in Settings
.envYour keys in plain text, only on a computer with no keychain Ixel can use (a Linux server, say), or one that refuses to keep Ixel's item. Health says so. Keys an earlier version of Ixel kept here move into keys.enc the first time it runs with a keychain, and so does a line you add by hand (Settings then lists it, so you can remove it)Until you remove them, or a keychain takes them
conversation.jsonYour last three ixel review questions and the final answers, so ixel review --continue can follow upA day: each one is deleted the next time Ixel runs after it's 24 hours old. Kept with Private on too, marked so a follow-up with Private off starts fresh
stats.jsonCounts: reviews, saves, what each month cost. No questions or answersKept
machines.json, machines_known_hostsYour machines and their pinned keysUntil you remove them
machines.logEvery connection, run, import and key change: which machine, its address, how it went, and the key fingerprints that show a server's key changed. Never a command's text30 days, in one file of at most 1 MB
update_check.json, device_keyWhen Ixel last checked for an update; a key for OpenClaw gatewaysKept
app.jsonYour Appearance choice: System, Light or DarkKept
The app window's browser storageThe Edge or Chrome profile Ixel's window uses on Windows and Linux (%LOCALAPPDATA%\IxelMAT\window, ~/.local/state/ixel-mat/window), or WebKit's storage for the Mac app. It holds the project folders you opened on the Board and the browser's own cache and settings, never a question or an answer. Ixel's own Linux window keeps it where WebKitGTK puts it instead (usually ~/.local/share/ixel and ~/.cache/ixel)Until you delete it (ixel forget; for the Linux window, delete those folders with Ixel closed, as ixel forget tells you)
Your Pictures folder, in IxelPictures you make (not what you asked for)Yours to keep or delete

Add-ons keep their own folder, ~/.ixel:

WhatWhat's in itHow long
addons/Cards the Agent wrote, each add-on's installed files, and state.json: which add-ons are on and which version of each card you approvedAn add-on's files until you switch it off; cards until you delete them
diagrams/The diagrams you made in Ask, as HTML filesYours to keep or delete

Handoff keeps each project's board in the project, in .handoff, which stays out of git:

WhatWhat's in itHow long
.handoff/board.dbEvery task with its notes, handoffs and reviews. Readable only by you (on Windows, the .handoff folder is set so only your account can open it)Until you delete the task (handoff delete) or clean out finished ones (handoff clean, or handoff keep 30 to have the board do it). Deleting overwrites the text in the file
.handoff/outputs/Answers, reviews and pictures from agent runs, a folder for each taskDeleted with their task
.handoff/worktrees/The copies the worker made for its tasks, on branches handoff/T-N, whose commit messages carry the task's titleUntil you remove them. Deleting a task tells you how

Never written to disk: the terminal chat's conversation and your earlier questions in it (gone when you leave), the app's conversations (kept in Ixel's memory so a reload brings them back, and gone when Ixel stops), pictures you attach (kept in memory for 30 minutes at most), and sound (not kept at all).

Things expire when Ixel runs: if you don't run it for a while, an old conversation.json waits until you next do. ixel forget deletes it at once.

What the AI programs keep

A program on your computer that Ixel hands a question to may save it in its own folders. Claude Code and Codex are told not to. Gemini CLI, Copilot and OpenCode have no such switch, so once each run is over, Ixel removes what that run saved, and nothing else: never a login, a setting, or a session of yours. Grok Build gets a home folder of its own for each run, holding only a copy of your computer's id (Grok Build's agent_id), your default model, and a setting that keeps your skills in ~/.agents out. Ixel deletes that folder afterwards. Your login stays where it is, and Grok Build refreshes it there as it always does. We checked this with Gemini CLI 0.62, Copilot 1.0.91, OpenCode 1.18 and 2.0, and Grok Build 1.0.46.

ProgramWhat it saves of a question, which Ixel removesWhat stays
Gemini CLIThe chat with your question and answer (in ~/.gemini/tmp, or ~/.cache/.gemini with its own sandbox on, on a Mac), its history folder, and the run's folder in its list of projectsA random installation id. With its sandbox on, Gemini CLI also puts the question on the command line of the copy it starts inside the sandbox, where other users of the computer can see it while it runs
CopilotThe session's folder in ~/.copilot/session-state, its rows and search index in ~/.copilot/session-store.db, and its log (Ixel gives each run its own). This needs Copilot 1.0.52 (May 2026) or later; Ixel says so if yours is olderIts settings file and a random device id
Grok BuildEverything it keeps of the run (the session with your question and answer), in the folder Ixel gave itNothing
OpenCodeThe run's sessions, with everything kept for them, in its database in ~/.local/share/opencodeIts log, about 11 KB a question with the folder, a session id and timings, never the question. OpenCode 2 can leave the question's text in its database file's free space, where nothing reads it, until later use writes over it

Removed rows are overwritten, not just marked free. If one of these programs is busy with its database at that moment, Ixel waits two seconds, then leaves it, and your answer isn't affected. An agent you point at a folder of your own (its workdir) is left alone, since your own sessions are kept there too; the folder Ixel gives Grok Build for each run is still deleted.

What Ixel strips or refuses

  • Photo details are removed. Every picture you attach is drawn again before it's sent, which leaves out what a camera writes into a photo, such as where it was taken.
  • Secrets in code are refused. Code or files that look like they hold a key or a token aren't sent, and Ixel says where they are. Handoff refuses them on the board too.
  • Subscription programs don't get your keys. The keys Ixel keeps aren't passed to Claude Code, Codex and the others, so your subscription is what gets used. The one exception is Gemini CLI: it gets your Google (Gemini) key while it isn't signed in another way.
  • No redirects. A key is sent only to the address you set, never on to wherever that address points.
  • A folder can't change Ixel's settings. Ixel ignores settings files in the folder you run it in, so a project you download can't point your keys elsewhere.

Honest limits

  • Your keychain gives your keys to programs you run. The key that opens your saved keys is in your keychain, and on Windows any program you run can ask Credential Manager for it, as can any program in your session once a Linux keyring is unlocked. A Mac asks you first. A copy of your home folder is only as safe as your keychain's password, and a Linux keyring with a blank password protects nothing. On a computer with no keychain Ixel can use, your keys are in plain text, in a file only you can read.
  • On Windows and Linux, the app window is Edge or Chrome. Ixel turns off its background services, but two requests to Google are left (above), and Edge's own Microsoft services follow Microsoft's settings, not Ixel's.
  • A question typed on the command line (ixel review "…") is saved in your shell's history, like any command. Type it in the ixel chat to avoid that.
  • Private covers your questions, not everything. Pictures you make with ixel image still go to xAI or OpenAI, the Board's Review… and Fix… send work to the agent you pick, and a question asked through the plugin came from that app's model in the first place. Opening Settings and checking your models still send your keys to their companies, with no question. An ixel chat already open when you turn Private on doesn't follow it until you start it again.
  • Private trusts the servers you add. It can't see what a server on your network does with a question, so a server that passes questions on to a company still counts as yours. Private has the details.
  • Your own command-line agents. One you add yourself gets the question on stdin, unless its prompt_via says "flag", "arg" or "auto": then it's on the program's command line, where other programs on the computer can read it.
  • Deleting isn't shredding. Ixel and Handoff overwrite deleted text inside their own files, but a disk can keep old copies of a file it rewrote (most SSDs do), and so do backups and synced folders.

Leave as little as possible

  1. Use local models, with Private on

    For anything that mustn't leave your computer, add a local model and turn on Private in Settings, under Asking. Then only your own models answer, and nothing is sent if none of them can.

  2. Ask in the chat or the app

    The chat and the app keep their conversations in memory only. ixel review "…" keeps your last three questions in conversation.json for a day and puts the question in your shell's history.

  3. Clear what's kept

    Close Ixel, then run:

    ixel forget

    It deletes conversation.json, the Machines log and the app window's browser storage, and says what it removed. In the app, Forget at the bottom of Settings clears the first two. Ixel's own Linux window keeps its storage in ~/.local/share/ixel and ~/.cache/ixel, which ixel forget names but doesn't delete: delete those folders too, with Ixel closed.

    For a project's board, handoff delete T-N removes one task with its results, and handoff clean removes finished tasks that haven't changed in 30 days (--older-than DAYS to change that), after listing them and asking. handoff keep 30 has the board do that by itself. Deleting the project's .handoff folder removes everything.

    If you uninstall Ixel, also remove the keychain item named Ixel (account keys): in Keychain Access on a Mac, Credential Manager on Windows (Windows Credentials), or your keyring app on Linux.

  4. Turn off the update check

    Add this to config.toml, and run ixel update yourself now and then:

    [updates]
    check = false

This website

ixelai.com has no cookies, no analytics, no tracking, no forms and no accounts. Its fonts, scripts and pictures are served from the site itself, and each page's security policy stops it from connecting anywhere else. These docs read your browser's platform, in the page only, to show your system's steps first; nothing is stored or sent. GitHub Pages hosts the site and, like any web host, logs visitors' IP addresses for security; that log is GitHub's, not ours.

Found something that doesn't match this page? Email openixel.ai@proton.me.