Skip to content

Install

Set up ByteAsk with one prompt in your coding agent: Context, Cloud Agent or the ByteAsk CLI, whichever you choose. Your agent tells you the cost and shows you every change first.

From your coding agent

Copy the prompt and paste it into your agent. It points your agent at ByteAsk’s setup guide and asks it to tell you the cost before anything paid, and to show you every change before it makes it.

Set up ByteAsk (https://byteask.ai) for me.

Here are your instructions: https://byteask.ai/agent-guide.txt
Read them in full and follow them step by step.
Ask me what you need in one message, then do the whole setup yourself.
Never spend money without asking me first.
Manual setup
Works with
  • Claude Code
  • Codex
  • Cursor
  • Windsurf
  • VS Code

What your agent does

It looks first
Which agent, system and tools you have, and whether ByteAsk is already set up, with read-only commands. Nothing is installed yet.
It asks one question
Which products you want. Context is selected; Cloud Agent and the ByteAsk CLI are yours to add.
It tells you the cost
$0 to set up on the Free plan. Using the CLI and the Cloud Agent later draws usage credit. See Pricing.
It installs with one script
byteask.ai/install.sh, checked against its pinned SHA-256. A dry run shows every change first. It writes only under ~/.byteask and your agent’s own configuration.
It signs you in
A few short approvals in your browser, all with one ByteAsk account.
It checks and hands over
It verifies each part, then shows you what to do next and how to remove everything (sh install.sh uninstall).

Manual setup

Prefer to do it by hand? Each product’s reference:

Every agent
The same setup by hand for Claude Code, Codex, Cursor and other MCP clients, and how to remove it: Set up with your coding agent in the docs.
Context
Step by step in Claude Code, with every command, in the Context setup guide.
ByteAsk CLI
The install commands for Linux, macOS, Windows, pip, uv and npm are in the docs. What the CLI installer writes is below.
Cloud Agent
Nothing to install on your machine: set it up in the browser at app.byteask.ai.
Editors
Plugins for VS Code, Cursor, JetBrains, Neovim and Emacs drive the CLI. See Editors.
The guide
The exact steps your agent follows, for every agent: byteask.ai/agent-guide.txt.

What the CLI installer does

The ByteAsk CLI part of setup runs the CLI installer, code.byteask.ai/install.sh, after checking its pinned SHA-256. On native Windows the setup guide does not apply yet: install the CLI by hand with install.ps1 (the docs). Read off the two live scripts. Nothing here is inferred.

Linux and macOS, install.sh

Platforms
Linux and macOS, on x86_64 or arm64. Any other uname exits with an error before anything is downloaded. On Windows the script tells you to use WSL, or use the PowerShell installer below.
Install directory
Your PREFIX if you set one, otherwise /usr/local/bin when that is already writable by you, otherwise ~/.local/bin. The script never elevates to create it.
What lands there
Two executables, byteask-engine and the byteask launcher (a small POSIX shell script), plus a .engine-release marker file recording which release asset produced the engine, so a later update can skip a download it does not need.
Directories created
The install directory, ~/.byteask (or BYTEASK_HOME), and ~/.byteask/skills/compress and ~/.byteask/skills/terse for the two bundled skills.
Files in ~/.byteask
A gateway file, a config.toml, a downloaded models-catalog.json, the own-key sidecar and its translator modules, a SKILL.md per skill, and a one-shot referral file if you passed --ref=. An existing login token is read out of the old config.toml and written back into the new one, so an update does not sign you out.
Shell startup files
Not touched. The script does not append to .bashrc, .zshrc, or any other startup file. If the install directory is not already on your PATH it prints the export line and leaves you to add it.
~/.gdbinit
One block is appended, wrapped in marker comments and skipped if the marker is already present, so re-running does not duplicate it. The block sources ~/.byteask/byteask_gdb_bridge.py when that file exists, which is what defines the byteask-bridge gdb command. The block's own comment says nothing is armed until you run that command, and that you disable it by deleting the block or running byteask --uninstall-gdb-bridge.
sudo
Two best-effort package installs, each only when the tool is missing. On Linux: bubblewrap (the sandbox the engine uses for model-run shell commands) and ripgrep, through whichever of apt-get, dnf, yum, pacman, zypper or apk is present, prefixed with sudo when you are not root. On macOS bubblewrap is skipped entirely and ripgrep goes through brew, with no sudo. Every one of those branches swallows its exit code, so a failure is never fatal: the script prints a note and carries on.
Checksums
Before each candidate download the script fetches SHA256SUMS from the same source. If the manifest lists that asset and the digest disagrees, the file is deleted and refused outright. If the manifest is unreachable or does not list the asset, the install continues with a printed note. Set BYTEASK_REQUIRE_SHA256=1 to make a missing entry fatal as well. Either way every download must also clear a size floor of roughly 20 MB and an ELF or Mach-O magic-byte check.
Where the engine comes from
The release assets at github.com/ByteAsk/cli first, with code.byteask.ai as the fallback if that is unreachable. It walks a compressor ladder, .zst then .xz then .gz then the uncompressed asset, using only the decompressors actually on your PATH. The byteask launcher script itself comes from BUNDLE_URL, which is code.byteask.ai unless you override it.
Install ping
One short request to <BUNDLE_URL>/byteask/dl/<os>-<arch>, code.byteask.ai by default, capped at 3 seconds and ignoring failure. The script's comment calls it the per-platform install signal.
Environment overrides
PREFIX, BYTEASK_HOME, ENGINE_URL, BUNDLE_URL, GATEWAY_URL, MODEL (the model id written into config.toml), BYTEASK_REQUIRE_SHA256, BYTEASK_FORCE_ENGINE_DOWNLOAD, and BYTEASK_REF. All are read from the environment with a default, so exporting one before the pipe is enough.

Windows, install.ps1

Platforms
64-bit Windows only. The script errors out on a 32-bit OS. It runs under the built-in Windows PowerShell 5.1 or pwsh 7.
Install directory
Your PREFIX if set, otherwise %LOCALAPPDATA%\Programs\ByteAsk.
What lands there
byteask-engine.exe, a byteask.ps1 wrapper, a byteask.cmd shim so the command works from any shell, and the same .engine-release marker.
PATH
This one does edit your environment: if the install directory is not already in your user Path, it is appended to the registry-backed user PATH. The script tells you to open a new terminal to pick it up.
Config
%USERPROFILE%\.byteask (or BYTEASK_HOME) gets the same set of files as the Unix installer: gateway, config.toml, model catalog, the own-key sidecar and translators, and the two skills. An existing login token is carried across.
Elevation
None. Ripgrep is attempted through winget, scoop or choco if one of them is on PATH, inside a try/catch, and a failure only prints a note. There is no ~/.gdbinit step on Windows.
Checksums
The same policy as the shell installer: a listed digest that disagrees is refused, an absent manifest is a printed note, and BYTEASK_REQUIRE_SHA256=1 makes absence fatal. Downloads must also clear the 20 MB floor and an MZ magic-byte check.

Uninstall the CLI

If your agent set it up, ask it to run the uninstall step (sh install.sh uninstall --components cli). By hand, delete what the CLI installer created:

Executables
Delete byteask, byteask-engine and .engine-release from your install directory: PREFIX, otherwise /usr/local/bin or ~/.local/bin.
~/.byteask
Delete it (or BYTEASK_HOME): config, model catalog, sidecar files and skills.
~/.gdbinit
Delete the ByteAsk block, from the marker comment to the matching end marker.
Windows
Delete %LOCALAPPDATA%\Programs\ByteAsk (or your PREFIX) and %USERPROFILE%\.byteask, then remove the install directory from your user Path.
Left behind
bubblewrap and ripgrep stay installed; remove them with your package manager. Shell startup files on Unix are never touched.

Editors

The CLI does the work; these connectors bring it into your editor.