@krmxd/onegpt
v0.2.5-fix-beta
Published
OGPT - AI coding assistant for the terminal with a built-in local engine
Maintainers
Readme
OGPT (onegpt)
AI coding assistant for your terminal with a built-in local engine. The Node.js build of OGPT - same commands, tools and session format as the Python version.
Install
# universal launcher-installer (repo folder):
sh ../install.sh # or: sh install.sh at the repo root# zero-install (no PATH changes needed):
npx @krmxd/onegpt
# permanent `ogpt` command:
npm i -g @krmxd/onegpt
ogpt # alias: ogpt-nodeUsed
npm i @krmxd/onegptwithout-g? npm doesn't put local bins on your PATH - launch withnpx @krmxd/onegpt(OGPT prints this tip too).First launch auto-installs anything missing - including Ollama - hidden behind one line:
Installing packages...Install/uninstall notes: postinstall places
ogpt+ogpt-nodelinks in$PREFIX/bin(Termux),~/.local/bin, or/usr/local/bin- whichever works - so the command exists even for local (no-g) installs.npm rm -gmay leave those two symlinks behind (npm skips uninstall scripts); delete them by hand if you care.
Troubleshooting
ModuleNotFoundError: No module named 'ogpt' from a path like
.../usr/bin/ogpt - that is the Python build's launcher, not this
package. An old pip install is shadowing ogpt. Fix:
pip uninstall ogpt && rm -f $PREFIX/bin/ogpt
npm i -g @krmxd/onegpt@latest --forceIf ogpt is still grabbed by something else, the Node build is also
installed as ogpt-node (same app, unshadowable name).
Requirements
- Node.js >= 18
- Ollama running locally (
ollama serve) with a model pulled
Highlights
- 24 built-in tools: read/write/edit files, run Python & Node code, install packages (pip/npm), call HTTP APIs, git, web search & fetch
- Self-installing: missing packages - including Ollama itself if the
device doesn't have it - install quietly behind a single
Installing packages...line; never raw installer logs (OGPT_NO_AUTO_INSTALL=1disables) - Smart Project Mode: "build me a portfolio website" → it plans, writes complete HTML/CSS/JS (never stubs), verifies the files exist and only then reports done - lazy one-file answers get auto-corrected
- Permission prompts only for genuinely risky actions (shell, installs, git commit) - everything else just works
- Token dashboard with live tok/s charts at
http://127.0.0.1:8756(/dash) - Session history shared with the Python OGPT build (
/save,/load,/sessions)- resume where you left off:
ogpt --continue(or--session <id>)
- resume where you left off:
- Anti-sloth guardrails: tiny local models are kept on-task until every requested file actually exists
Useful commands
| Command | What it does |
| ------- | ------------ |
| /help | all commands |
| /model | switch model |
| /tools | list tools |
| /auto | toggle auto-approve |
| /dash | token dashboard |
| /new | fresh chat |
| /export [md|json] [file] | save the chat as Markdown or JSON |
| /todos | show the current todo list |
CLI flags: ogpt -m <model> · -c/--continue · --session <id> ·
--auto · --no-stream.
Config lives at ~/.config/ogpt/config.json.
