build
Makes the change. Every command and file edit stops at a gate first.
asg wants to run go test ./internal/... y run a always here g everywhere n block
The coding agent that asks first.
ASG runs in your terminal, works with local Ollama models out of the box, and waits for your yes before it runs a command or changes a file.
Free and open source · macOS · Windows · Linux
Shift + Tab cycles them
Makes the change. Every command and file edit stops at a gate first.
asg wants to run go test ./internal/... y run a always here g everywhere n block
Only reads. Hands back a numbered plan and is never given a tool that can change your project.
plan · 3 steps
1. add a --json flag to cmd/report
2. encode rows with encoding/json
3. table test for both outputs
Answers questions from the actual files in your repo, not from what a model remembers.
› where do we retry failed uploads? read internal/sync/upload.go retryUpload, line 88: 3 tries, backoff ×2
Treats your message as an objective. The harness splits it into steps, checks each one and runs your tests.
/loop add rate limiting to the API step 2/5 · review passed verify · go test ./... ok
/loop <objective>
Small local models do their best work with a short, fresh context. So each step gets one, every step is reviewed, and nothing is called done until your tests pass.
Runs save after every step. /loop resume picks up after a restart, a question or Ctrl + C.
What's in it
edit internal/api/limit.go +24 −3 y a g n · ctrl+y for YOLO
a allows that kind of call in this project. /permissions takes it back.
Ollama
first
Anthropic, OpenAI, OpenRouter, Groq, LM Studio or anything OpenAI-compatible. One flag switches, and ASG remembers.
asg -provider anthropic
×4
After every answer ASG reads its work back four times, each from a different angle, and fixes what it finds.
Ctrl + T cycles them. Docs screenshots every edit. Cyber confirms scope first and ends with a report.
Kept in .asg/work/ on every build and loop run.
Notes live in ~/.asg/memory/ as plain Markdown. Install SKILL.md skills from GitHub, a URL or a folder.
asg skill add owner/repo
Paste a screenshot with Ctrl + V. Have replies read aloud by the model, Ollama, OpenAI TTS or your OS voice.
The desktop app and setup wizard
Every button below fetches the newest release straight from GitHub.
asg-setup_windows_amd64.exe
Download for Windowsasg-setup_darwin_universal.dmg
Download for macOSUniversal: Apple silicon and Intel
asg-setup_linux_amd64
Download for LinuxARM64 build · then chmod +x and run it
Just the terminal app
The script picks your OS and CPU, checks the SHA-256 checksum and puts asg on your path. asg update upgrades in place.
For local models, install Ollama and pull one: ollama pull qwen3.5:9b. For hosted models an API key is enough.
bash / zsh
curl -fsSL https://raw.githubusercontent.com/kxngHADES/asg-harness/main/install.sh | shPowerShell
irm https://raw.githubusercontent.com/kxngHADES/asg-harness/main/install.ps1 | iexany OS with Go
go install github.com/kxngHADES/asg-harness/cmd/asg@latest$ cd your-project
$ asg