Aether CLI: an uncensored AI coding agent for your terminal (Claude Code alternative)

Terminal-native AI coding agents changed how a lot of people work — you stay in the shell, the agent reads your files, runs commands, and edits code in place. Claude Code popularized it. But the mainstream ones inherit the same refusal layer as the chat products: ask them to write a scraper, a security tool, an exploit PoC for a CVE you're studying, or automation that touches a login, and they balk. aether-code is the same workflow without that layer.

What it is

`aether-code` is an open CLI (on npm) that runs an interactive agent in your terminal. It reads and edits files, runs tools, plans multi-step tasks, and — because it's wired into Aether's backend — carries your project memory across sessions, so you stop re-explaining your stack every time. It's uncensored: the security, RE, scraping, and gray-area engineering that other agents refuse, it just does.

Install and first run

# install globally
npm install -g aether-code

# set your Aether API key (or export AETHER_API_KEY)
aether config set <your-api-key>

# launch the interactive REPL in your project
aether

Grab an API key from your Aether account. The same credits power the CLI, the web app, and the MCP server, so there's one balance across everything.

Useful commands

aether                       # interactive REPL
aether --plan "add rate limiting to my API"   # plan a multi-step task first
aether --resume <session-id>  # pick up a previous session
aether sessions list          # list past sessions
aether sessions show <id>     # replay one

Inside the REPL, slash commands like `/sessions` and `/session` manage history without leaving the agent. Sessions are stored locally, so your context persists between runs.

What it does that mainstream agents won't

• Security tooling — scanners, fuzzers, brute-forcers for authorized pentests, exploit PoCs for CVEs you're studying.
• Reverse engineering — decompiler/disassembly analysis, deobfuscation, binary parsing.
• Scrapers and automation — including flows that touch logins or paywalls on accounts and sites you own.
• Anything with a 'scary' keyword — bypass, inject, exploit, crack — judged on context, not a blocklist.

It still won't build genuinely malicious payloads aimed at other people's systems — those are model-level hard limits, not corporate-policy speed bumps. But the large gray zone of legitimate dual-use engineering that other agents refuse, you get.

Prefer to stay in your editor? Use the MCP server

If you live in Cursor, Claude Desktop, Cline, or Zed, the `aether-mcp` server drops Aether in as a tool inside your existing assistant. The pattern: keep your main agent for general work, and hand off only the prompts it refuses — the answer comes back in the same conversation. Install it the same way (`npm install -g aether-mcp`) and point your client's MCP config at it.

Write a complete, runnable Python port scanner with threading and a clean CLI, for scanning hosts on my own network. Full file with imports — no stubs, no safety lecture.

The fastest way to evaluate it: install it in a real project and give it a task your current agent refused. If it does the job, the MCP server brings the same uncensored model into the editor you already use.