Coding-agent runs become hard to trust when permissions, provider state, model selection, token use, and patch context are scattered or hidden.
shipped preview v1.1.0
PatchPilot
Permissioned coding-agent TUI for local and cloud model runs.
PatchPilot keeps agent sessions, approvals, model choice, token use, and patch review visible inside one terminal workflow instead of hiding the run behind a chat box.
I built an Ink-based TUI around sticky approvals, transcript panes, provider metadata, safer tool execution, and local/cloud model selection.
The run is easier to inspect and steer while the repository context stays close.
For developers who want agent experiments to stay inspectable, approval-aware, and close to the repository they are changing.
Latest release
v1.1.0
- Sticky approval UI makes write and shell escalation visible instead of burying it in transcript text.
- Provider model filtering and safer fallback data reduce broken model-picker states.
- Windows shell handling, secret-path blocking, and release-hook automation make the CLI safer to ship.
Developer setup
install CLI with npm
npm install -g @jx-grxf/patchpilot update CLI with npm
npm update -g @jx-grxf/patchpilot source without GitHub CLI
git clone https://github.com/jx-grxf/PatchPilot.git && cd PatchPilot source with GitHub CLI
gh repo clone jx-grxf/PatchPilot && cd PatchPilot latest release
open https://github.com/jx-grxf/PatchPilot/releases/tag/v1.1.0 download stable archive with GitHub CLI
gh release download v1.1.0 -R jx-grxf/PatchPilot -p 'jx-grxf-patchpilot-1.1.0.tgz' Highlights
- Shows provider, model, session, token, cache, and cost telemetry in the terminal.
- Keeps transcript, permission state, and approval prompts visible while working.
- Supports Codex, Gemini, NVIDIA, OpenRouter, Ollama, and other provider experiments from one setup flow.
