Plugins and skills
Give the agent the tools and workflows your work needs. Install plugins and reusable skills, connect MCP services, and use them directly from the conversation.
Const Harness
From coding to office work, Const Harness connects to your local files, tools, and browser to help you execute, organize, verify, and turn ideas into deliverable results.
Const Harness makes work easier.
Every task, delivered with quality
ConstHarnessisadesktopAIworkbenchfordevelopers,researchers,andcreators.Bringyourprojects,models,andtoolstogethertoturnanideaintoworkingcode,usefulresearch,andfinishedcreativework.
The best ways to use Const Harness
Native computer use
Work beyond the browser. Native computer-use helpers let the agent inspect and interact with desktop apps on Windows, macOS, and Linux.
Const Harness makes work easier.
Browser control
Research, navigate, and test in a real browser inside the workbench. Inspect page structure, capture screenshots, and check your site while you build.
Create images, video, and audio
Combine creative skills with your configured image, video, and speech providers. Build presentations and motion projects, then keep the outputs with the rest of your work.
A terminal for real work
Run commands, builds, and long jobs without leaving the workbench. Follow background tasks and subagents, inspect their output, and stop them when needed.
pnpm --filter @const/harness-desktop dev
Use Const Harness wherever you work
Const Harness makes work easier.
Desktop
A native workbench for Windows, macOS, and Linux, with the chat, terminal, browser, diff, and review panes in one window.
Download$ pnpm check-types
> tsc --noEmit
β 0 errors
$ pnpm build
β Build completed
$ βTerminal
Use the built-in terminal alongside your project, conversation, and review pane. Keep your existing scripts and development tools in the same workflow.
Explore the terminalDaily project checkEvery day at 09:00
Last run completed
Scheduled tasks
Set recurring research, reports, and project checks. Runs execute on your computer while the app is running, with results saved to a conversation.
How scheduling worksConst Harness vs. a chat-only workflow
This compares two ways of working, not individual vendors. Chat products with agent tools can also provide execution and integrations.
Get started with Const Harness
One workbench. Choose how you connect your models.
Desktop workspace
Start here
Install the app, sign in with Const Accounts, and open your project.
Get the desktop app- Windows, macOS, and Linux
- Project conversations and file review
- Built-in terminal and browser
Your own provider
Your API key
Connect a supported provider and choose the models for your work.
Explore model setup- OpenAI and Anthropic protocols
- Custom compatible endpoints
- Provider usage billed separately
Const Router
One gateway
Access the models available through your Const Router account.
Open Const Router- Models served through Const Router
- Choose a model in the workbench
- Quotas depend on your account
Model availability, quotas, and provider charges depend on your configuration and account.
Experience liftoff with Const Harness
The modern agentic workbench for developers. Ready for your workflow.
Frequently asked questions
Install a desktop build or build from source, sign in through Const Accounts, and choose a project folder. Configure Const Router or a supported model provider, then describe the task and choose its permission mode.
No. The workbench also supports research, data analysis, documents, presentations, and creative workflows through skills and plugins. Image, video, and speech workflows require the corresponding configured provider or tool.
A skill is a reusable set of instructions for a type of work. A plugin packages capabilities such as skills and service connections. MCP connections give the agent access to tools exposed by a configured server.
You choose the permission mode. Review mode requests approval for actions; ask-before-changes allows file edits inside the workspace while asking for shell execution and writes outside it; full access is more permissive. Conversation history is stored locally, but prompts and relevant context are sent to the configured model provider. Connected tools can also send data to their services.
Yes. Set instructions, a workspace, a model, permissions, and a recurring schedule. Runs are recorded in task history and results are saved to a conversation. Execution is local: keep the desktop app running and your computer awake.
The desktop project includes packaging for Windows, macOS, and Linux; available binaries depend on the published release. Configure Const Router or supported OpenAI- and Anthropic-compatible providers. The standalone CLI in this repository is still a scaffold; use the desktop's built-in terminal for execution.