Promptline

Single-binary Go terminal for LLM chats, built-in tools, no system execution calls, ultra portable.

We are hackers. This is our straight‑shot terminal client for OpenAI‑compatible LLMs: batch or streaming, tool calls, sane permissions, no deps, no system exec.

Terminal interface for streaming chat with OpenAI-compatible APIs. Written in Go. No exec. No deps. Just a single binary that works over SSH.

Get started → build, run, configure and hack in minutes.

What it does

Streams OpenAI-compatible chat directly to your terminal.
Supports tool-calling with strict allow/ask/deny permissions.
Runs in interactive mode or batch/pipe mode.
Ships with Go-native tools, never calls system binaries.

Downloads

Latest releases on dyne/promptline - direct links:

Linux macOS Windows

Features

Streaming: real-time response chunks in the console.
Permissions: allow/ask/deny per tool, default is ask.
Batch mode: pipe queries in, get output out.
Portable: single binary, SSH-friendly, no GUI dependencies.
Fully configurable: one JSON in current folder.
Zero dependency: add tools in Go, no shelling out.

Quick start

Have Go installed to build on your machine

  make build

Or just download the latest release and run

  ./promptline

Batch mode:

  echo "write hello world in hello.txt" | ./promptline -

Tools included

Core

get_current_datetime, read_file, write_file, ls

File operations

cat, cp, mv, rm, ln, touch, truncate, readlink, realpath

Directory operations

mkdir, pwd, dirname, basename

Text processing

grep, head, tail, sort, uniq, wc, tr, tee, comm, strings, more

File viewing / analysis

hexdump, cmp, md5sum, shasum, base64

System info

uname, hostname, uptime, free, df, du, ps, pidof, id

Misc safe

echo, seq, printenv, tty, which, mkfifo, mktemp, find, chmod, date

Technical details

API: OpenAI-compatible (api_url configurable).
No exec: tools are native (u-root), no exec.Command.
Portable: just one executable, available for GNU/Linux, macOS and Win
Secure: defends host from malevolent LLM agents.
Permissions: allow/ask/deny per tool (default ask).
Limits: file size, directory depth, directory entries.
Throttling limits: per-tool and global, with cooldowns.
Timeouts: global default and per-tool overrides.

Config

config.json or environment variables.

  {
    "api_key": "sk-...",
    "api_url": "https://api.openai.com/v1",
    "model": "gpt-4o-mini",
    "tools": {
      "allow": ["read_file", "ls"],
      "ask": ["write_file"]
    },
    "tool_limits": {
      "max_file_size_bytes": 10485760,
      "max_directory_depth": 8,
      "max_directory_entries": 2000
    },
    "tool_rate_limits": {
      "default_per_minute": 60,
      "per_tool": {},
      "cooldown_seconds": {}
    },
    "tool_timeouts": {
      "default_seconds": 0,
      "per_tool_seconds": {}
    }
  }

License

GNU AGPLv3+ — free as in freedom.

🐃 Sourcecode

Promptline is 100% free and open source software running as a single binary, written in Go language. Our source-code is available on dyne/promptline.

🧙 May the source be always with you!
🧑‍🏫 for professional support contact: The Forkbomb Company