Bring the project.
Open local or SSH workspaces. Keep terminals, project context, files, and Git close enough to use together.
Local-first · Open source
A workbench
for the agent era.
Run coding agents. Supervise the changes.
Keep the whole project in view.
PacketBench brings terminals, API conversations, planning, and review into one native desktop environment.
P> Built by an infrastructure engineer. Designed for
the work around the agent.
THE WORKBENCH01 / LOCAL DESKTOP
PacketBenchYOUR WORK, IN CONTEXT
◇
01 / PLANSet the direction
>_
02 / RUNBring your agents
±
03 / REVIEWInspect the changes
↻
04 / CONTINUEKeep your bearings
CLI + API · LOCAL + SSHWORKFLOW ILLUSTRATION
01Native desktopTauri + Rust
02Multiple agent pathsTerminal CLIs + API agents
03Inspectable sourceApache-2.0
01 / Inside the workbench
The project stays
in the picture.
Move from a conversation to a supervised run, with the tools and controls close to the work.
Plan first. Launch when you’re ready.
Choose local or SSH targets, define execution supervision, and use isolated worktrees with optional review gates.
View full capture ↗
A place to direct the work.
Start API-agent conversations against a project, choose the provider and execution posture, and keep conversation context visible.
View full capture ↗
Make the workbench yours.
Configure terminal agents, execution hosts, and project rules. PTY-first Workspaces keep your terminal workflow in a persistent pane layout.
View full capture ↗
02 / One continuous workflow
From intent
to inspected work.
The useful part is everything that connects the prompt to the project.
Choose how the agent works.
Run familiar terminal CLIs or API-agent conversations. Select the provider and permissions that fit the job.
Organize the larger job.
Track issues, plan Flights, and launch supervised attempts against isolated worktrees.
Review, then continue.
Inspect pending edits, accept changes deliberately, and keep notes and history available for the next session.
03 / Your agent stack
Bring the tools
you work with.
Choose the terminal workflow or API path that suits your project. Provider credentials and capabilities depend on your configuration.
TERMINAL CLIS
PacketCode / Claude Code / Codex / OpenCode
API + LOCAL MODELS
Claude / OpenAI / MiniMax / OpenRouter / Ollama
MORE CONFIGURABLE ENDPOINTS
See the current provider matrix
Sugar / Gemini / xAI / OpenAI-compatible services
04 / Engineering with boundaries
Keep authority
visible.
Automation earns trust through controls you can see and decisions you can make.
HUMAN IN CONTROL
Approve the action.
Explicit permission prompts and project trust put tool access in front of the person directing the work.
Review the change.
Inspect pending edits and use review surfaces before accepting work into the project.
Bound the work.
Configure spending limits and supervised attempts. Bounded autonomy is an explicit choice.
Preserve context.
Persisted conversations, workspace layouts, notes, and history keep the next step understandable.
05 / Built in the open
Inspect the work.
Follow what’s next.
Current source version0.14.9ACTIVE DEVELOPMENT
A working desktop project, still evolving.
The recorded Windows build has been installed and launched. Interactive Workspace use and real-account provider acceptance are still being completed. The repository keeps build evidence, current limits, and planned work visible.
A few useful details
Before you
open the source.
Is this a browser app?
PacketBench is a native desktop application built with Tauri and Rust. This page introduces the product; the repository contains the source, setup instructions, and build records.
Does it replace my coding agent?
It provides the environment around the agents you use: terminal panes, API conversations, project organization, permissions, and review. CLI and API paths remain distinct.
Can I work on remote projects?
SSH workspaces and remote execution paths are supported. Remote hosts require their own compatible tools and configuration. The separate Remote Agents relay feature remains gated; it is not a general hosted service.
Where should I start?
Start with the GitHub README for prerequisites and setup. Read the current handoff for acceptance status, and use the documentation to explore the workflow.
Your agents. Your projects. Your workbench.
Bring your agents.
Keep your bearings.
Explore PacketBench on GitHub
Open source · Apache-2.0 · Native desktop