devoholic.AGENT STUDIO
A writing partner. A research helper. A second pair of eyes on your code. Give your agent a purpose, connect its tools, and run it on your own computer.
A little structure for your next big idea.
“Help me turn my portfolio website idea into a realistic weekend plan.”
You don’t need to start with a framework.
Start with something you want help with.
Bring your goals and constraints. Get a plan you can actually start.
Build in the dashboard. Run on your computer. Follow the work from the first instruction to the finished answer.
Start with this template ↗See request and tool activity, explore formatted answers,
and export the work you want to keep.
✓ Request received on your computer
✓ Model response received
✓ Answer ready on the canvas
✳ Here’s a comparison, with trade-offs and a suggested next step.
The right starting point depends on how much time you have and how much you want to customise.
| Approach | Good for | Trade-off |
|---|---|---|
| Use a template | A first useful agent | Adapt the instructions |
| Customise a project | Your specific workflow | More setup choices |
| Connect tools | Working with your services | Configure and trust each tool |
Start small. Make one agent useful, then add capabilities.
Model waits and actual MCP tool events with an updated companion.
Readable reports, tables, code and basic flowcharts.
Copy an answer or export Markdown before leaving the chat.
The dashboard is where you build and connect.
Your local companion is where agents run and credentials are stored.
Build, chat and see status.
Agent projects · encrypted key vault
Runtime · connected tools
Local Ollama, cloud AI
and your approved MCP services.
Agent projects and the encrypted credential vault are stored on your paired computer. Using a cloud model or an external tool sends the data included in that request to that service. The platform sees account and device information and aggregate activity counts.
Windows setup prepares missing tools and a desktop shortcut. Open the companion, sign in, and return to your saved agents. The computer must be online and awake. Windows may request installer permission.
Use a built-in template, change its instructions,
then add the capabilities your task needs.
Loading starting points…
Templates provide editable instructions. External services aren’t preconnected. Bring and test the tools you want your agent to use.
You can start with a template and edit its purpose and instructions in the dashboard. The studio generates the agent project. Custom tools and advanced integrations can still need technical setup; they aren’t automatically configured for you.
A skill is a reusable instruction file, such as a writing style guide. MCP is a way to connect an agent to tools and services. Skills teach your preferences; tools let the agent do things. You can begin without connecting external tools.
The runtime, agent files and encrypted credential vault stay on the paired computer. Requests to a cloud model or external tool send their included data to that provider. The dashboard relays encrypted traffic and stores account details, device metadata and aggregate counts. Local execution alone is not a guarantee of offline or sandboxed operation.
Yes. The companion discovers available models in Ollama on your computer. You can also connect a supported cloud provider with your own API key. A fully offline workflow needs local models and local tools; setup and model downloads can require internet access.
Your computer must be awake and the companion must be running. If it goes offline, the dashboard cannot run agents on it and does not switch to server execution. Starting the same companion again reconnects your saved workspace.
Not in the current release. Local commands run with your operating-system permissions. Command MCP must be enabled explicitly on the companion. Review configurations and use tools you trust. General-purpose sandboxing and automatic approval gates for every tool action are not currently available.
One agent. One useful task. A good place to begin.
Create your first agent ↗Or find a template ↑