ByteAsk
ByteAsk is a terminal-native AI coding agent built for C and C++ development. It edits your repository from the terminal, then drives the real toolchain — sanitizers, your test suite, and any gdb session you already have open — and hands you a diff that survived the debugger.
Your prompts and code are processed to answer the request, then dropped. Nothing is stored, and your repositories are never used to train models. See the Privacy Policy.
Installation
Linux and macOS
curl -fsSL https://code.byteask.ai/install.sh | sh
Windows
irm https://code.byteask.ai/install.ps1 | iex
Python (pip / uv)
pip install byteask
# or
uv tool install byteask
Node (npm)
npm install -g @byteask/cli
Update
byteask --update
If you received a referral code, prefix the install command with BYTEASK_REF=<code>:
BYTEASK_REF=MYCODE curl -fsSL https://code.byteask.ai/install.sh | sh
Supported platforms
| Platform | Architecture |
|---|---|
| macOS | Apple Silicon (arm64) |
| Linux | x86_64, arm64 |
| Windows | x86_64 (native, no WSL required) |
Quick start
After installing, sign in with a magic link:
byteask login
Then run it inside any Git repository:
cd ~/your-project
byteask "find the memory leak in parser.cpp and fix it"
ByteAsk opens a full-screen TUI, reads your codebase, edits files, compiles, runs tests, and presents a diff for your review before making permanent changes (by default).
Non-interactive mode
For scripting and CI pipelines, use the exec subcommand:
byteask exec "add null-checks to all public API entry points"
Config file
ByteAsk reads ~/.byteask/config.toml on startup. Generate a starting config with:
byteask doctor
Key settings:
# Model to use
model = "gpt-5.4"
# How much the agent can do without asking
# "on-request" (default) | "untrusted" | "never"
approval_policy = "on-request"
# Web search capability
# "disabled" | "cached" (default) | "live"
web_search = "cached"
Override any key inline without editing the file:
byteask -c model="gpt-5.4" -c approval_policy="never" "refactor the logger"
Feature highlights
| Feature | Description |
|---|---|
| Interactive TUI | Full-screen terminal UI with streaming output, approval prompts, and session history |
| 18 C/C++ tools | Compiler Explorer, sanitizers, gdb, rr, perf, clang-tidy, and more — built in |
| Approval modes | Fine-grained control over what the agent can do autonomously |
| MCP servers | Connect any Model Context Protocol server |
| Session management | Persist, resume, fork, and archive conversations |
| Multimodal input | Attach images to any prompt |
| Non-interactive mode | Scripting-friendly exec subcommand with JSONL output |
| Sandboxing | Bubblewrap (Linux) / Seatbelt (macOS) filesystem isolation |
| Vim mode | Vim keybindings in the composer |
| Plans, usage & credit | Track usage, upgrade, and earn credit via /usage, /upgrade, /referral-code, /voucher |
For the full list, see Features. For the in-app commands (including the ByteAsk-specific /usage, /referral-code, /voucher, and /upgrade), see Slash Commands.