aicmd

Turn plain English into a shell command without leaving your terminal.

>>> aicmd <description of command...>
<ai-generated command>    <-- inserted into terminal, ready to execute

>>> aicmd spin down the app on 8080
pids=$(lsof -ti tcp:8080); [ -z "$pids" ] || kill $pids

>>> aicmd list js files recursively by date modified
find . -type f -name '*.js' -print0 | xargs -0 stat -f '%m %N' | sort -n

Install

curl -fsSL https://github.com/zazencodes/aicmd/raw/refs/heads/main/install.sh | sh

MacOS and Linux are supported.

Why aicmd

Stay in the terminal:

• Be faster.
• Use your mouse less.

Smart AI options:

• CLI Agents – AI that you already pay for.
• Direct API – Lower latency but higher cost per token.
• Local LLM – Completely free but less capable.

Created by ZazenCodes

How it works

aicmd is very simple. It finds the best terminal command for your prompt.

The generated command appears on a new line, ready to send. You can review, modify and run it as you like.

It's powered by the CLI that you already pay for: Codex, Claude Code or Antigravity, e.g. codex/gpt-5.6-terra/low

Or you can use the APIs directly. This incurs some API costs but often has lower latency.

Under the hood, your aicmd request is dynamically injected into the following prompt, which is passed to the underlying LLM:

You convert natural language into a single shell command.

Environment:
- OS: <os>
- arch: <arch>
- shell: <shell>
- cwd: <cwd>

Rules:
- Output ONE single-line shell command and nothing else.
- No markdown, no backticks, no explanation, no leading `$` prompt.
- The command must run as-is in the environment above.

Request: <your request>

Usage

aicmd <description of command...>

Flags:

The generated command is placed at your prompt for review. When output is redirected, e.g. with a pipe command, the aicmd result is printed instead of being inserted into the terminal.

See the usage guide for a more detailed guide on input forms, flag placement, terminal injection, pipelines, provider errors, and safety behavior.

Configuration

The installer creates a documented starting config in ~/.config/aicmd.toml

model = "openai-api/gpt-5.6-luna/none"   # provider/model/effort
clipboard = false                   # also copy the command to the clipboard

Manage it with the config subcommand:

aicmd config list                            show all settings
aicmd config get model                       print one setting
aicmd config set model codex/gpt-5.6-terra/medium
aicmd config set clipboard on

API-backed providers use environment variables for credentials:

export OPENAI_API_KEY=...
aicmd -m openai-api/gpt-5.6/low <your request>

export GEMINI_API_KEY=...
aicmd -m gemini-api/gemini-3.7-flash/low <your request>

export ANTHROPIC_API_KEY=...
aicmd -m claude-api/claude-sonnet-4-6/low <your request>

Local models are available through Ollama and llama.cpp:

ollama serve
aicmd -m ollama/gemma4:e2b/none <your request>

llama-server -hf ggml-org/gemma-3-4b-it-GGUF --alias gemma-3-4b-it
aicmd -m llama/gemma-3-4b-it/none <your request>

See the local LLM guide for server setup, endpoint
overrides, authentication, and provider differences.

Development

Go 1.24. API providers use the official OpenAI Go SDK, pinned to the final
release compatible with Go 1.24.

make build       # build ./aicmd
make test        # unit tests
make vet         # go vet
make build-all   # cross-compile all 4 targets into dist/

Releases are tag-triggered (.github/workflows/release.yml): pushing a v*
tag builds the 4 targets, packages aicmd_<os>_<arch>.tar.gz +
checksums.txt, and creates a GitHub release.

Documentation

• Requirements — supported platforms and CLI, hosted
  API, and local LLM inference options
• Usage guide — prompts, flags, output behavior, safety, and
  provider troubleshooting
• Configuration guide — settings, model specs,
  provider differences, clipboard support, and config file format
• Local LLMs — Ollama and llama.cpp setup and configuration
• Platform support — terminal injection behavior on
  macOS, Linux, and Windows
• Architecture decision records — lasting design
  decisions and repository guardrails
• Modular provider architecture — the
  implemented design specification for provider integrations

Release notes

Each GitHub release includes a
curated summary of user-visible changes. The Git history remains the detailed
engineering record.