Sarkar Deskmate is a passion project developed under ., born from a simple goal: staying in creative and developer flow even when I’m not sitting at my desk.
I didn’t just want remote control.
I wanted an assistant that understands what I’m working on and lets me keep building, without switching mental gears or remembering commands.
That idea became .
Deskmate is a local-first agent that lets you continue working on your laptop from anywhere using natural language. It handles real, everyday “off-the-desk” moments like:
* Running a Docker build and push while you’re grabbing coffee
* Triggering tests or notebooks from bed
* Sending VPN-only emails or handling internal operations remotely
* Write code, run tests, or monitor running jobs and services remotely
* Take screenshots or join meetings even when you’re away from your machine
* Granting limited, time-bound access to your laptop when you’re not around
A simple example:
```
You: “What’s using port 3000?”
Deskmate: “node (PID 12345) – your Next.js dev server”
You: “Kill it and start the production build”
Done.
```
With Deskmate, there are no commands to memorize, no context switching, and no broken flow.
Just intent.
Under the hood
* A local agent runs on your laptop under your user account
* Instructions are interpreted and planned locally, with awareness of context
* Actions execute safely on your machine
* You get feedback through logs, screenshots, and confirmations
Building on ideas from gen-shell, Deskmate goes further by introducing agents, workflows, approvals, and remote execution, so your creative or developer mode continues even when you step away from your desk.
Current stack
* Claude Agent SDK for agentic execution
* MCP for Claude Desktop integration
* Telegram for quick remote interaction
* TypeScript, running as a macOS background service
Planning to open this up for open source soon and would love to collaborate with others who are interested in:
* Making background execution reliable across always-on machines and sleeping laptops
* Migrating the agent runtime to OpenCode or other open-source alternatives
Sarkar Deskmate is a passion project developed under ., born from a simple goal: staying in creative and developer flow even when I’m not sitting at my desk.
I didn’t just want remote control.
I wanted an assistant that understands what I’m working on and lets me keep building, without switching mental gears or remembering commands.
That idea became .
Deskmate is a local-first agent that lets you continue working on your laptop from anywhere using natural language. It handles real, everyday “off-the-desk” moments like:
* Running a Docker build and push while you’re grabbing coffee
* Triggering tests or notebooks from bed
* Sending VPN-only emails or handling internal operations remotely
* Write code, run tests, or monitor running jobs and services remotely
* Take screenshots or join meetings even when you’re away from your machine
* Granting limited, time-bound access to your laptop when you’re not around
A simple example:
```
You: “What’s using port 3000?”
Deskmate: “node (PID 12345) – your Next.js dev server”
You: “Kill it and start the production build”
Done.
```
With Deskmate, there are no commands to memorize, no context switching, and no broken flow.
Just intent.
Under the hood
* A local agent runs on your laptop under your user account
* Instructions are interpreted and planned locally, with awareness of context
* Actions execute safely on your machine
* You get feedback through logs, screenshots, and confirmations
Building on ideas from gen-shell, Deskmate goes further by introducing agents, workflows, approvals, and remote execution, so your creative or developer mode continues even when you step away from your desk.
Current stack
* Claude Agent SDK for agentic execution
* MCP for Claude Desktop integration
* Telegram for quick remote interaction
* TypeScript, running as a macOS background service
Planning to open this up for open source soon and would love to collaborate with others who are interested in:
* Making background execution reliable across always-on machines and sleeping laptops
* Migrating the agent runtime to OpenCode or other open-source alternatives