AI agent package family for Sunder

AI agents that can use real tools.

Sunder Agent runs inside Sunder App. Add the model providers and tool packages you choose, then let the agent read selected files, edit projects, run commands, search the web, call MCP tools, remember context, delegate to subagents, use local models, or build Sunder packages with Agent Builder.

Install Sunder Agent Browse Agent Packages View Agent Source

Practical tools

What Agent can actually do.

Agent starts as chat, but the useful part is the tool layer you decide to install around it.

Work with files

Read, search, create, edit, and patch selected workspace files when file tools are installed and allowed.

Run commands

Use shell tools with a selected execution target for builds, scripts, tests, imports, exports, and local workflows.

Research the web

Search the web or fetch pages as markdown, text, or HTML for grounded project work.

Use MCP tools

Connect local or remote MCP servers so Agent can call external tools and services.

Remember context

Recall project notes, preferences, standing instructions, and facts that matter across sessions.

Delegate work

Use specialist subagents for focused tasks while the main agent coordinates the work.

Use local models

Run workflows against LM Studio models when you want a local model surface.

Build packages

Use Agent Builder to create, edit, build, and test Sunder packages from inside the app.

Agent Builder

Create a package by describing the tool you want.

Agent Builder is the crucial non-manual path: create a new Sunder package inside Sunder, then ask Agent Chat to build it. The agent can generate files, edit code, run commands, test builds, inspect errors, and keep iterating without you starting from a blank desktop app setup.

Make me a dashboard for my Minecraft server.
Make a panel that tracks invoices from this CSV folder.
Make a tool that resizes images in this folder.
Make a button that runs my backup script.
Make a package for my Home Assistant devices.
Make a local model experiment workspace.

Install flow

Start small. Add only the tools Agent should have.

Agent is modular because local tool access should be explicit. Start with the core package, add a model provider, then add file, shell, web, MCP, memory, execution, or Builder packages as needed.

Download Sunder

Install the desktop app where Agent runs.

Install Agent

Add sunder.package.agent from the Registry.

Choose a model

Use OpenAI, Anthropic, Gemini, LM Studio, or future providers.

Add tools

Allow files, shell, web, MCP, memory, subagents, execution, or Builder only when you need them.

Start working

Ask questions, edit projects, run tasks, use tools, or build a new Sunder package.

Registry install

sunder install sunder.package.agent

Plain answers

Direct answers for AI agent users.

Short definitions for people comparing chatbots, coding agents, local agents, MCP tools, and package builders.

What is Sunder Agent?

Sunder Agent is a first-party package family that adds AI chat sessions, model providers, file tools, shell tools, web tools, MCP, memory, subagents, execution targets, and Agent Builder to Sunder App.

How do I install Sunder Agent?

Install Sunder App, open the public Registry, then install the Agent core package. Add only the model provider, tool, workflow, execution, or Builder packages you need.

sunder install sunder.package.agent

Can Sunder Agent use local models?

Yes. The LM Studio provider connects Sunder Agent to local OpenAI-compatible LM Studio endpoints and discovers available local models.

What tools can Sunder Agent use?

Agent tool packages can add file read/write/edit/apply_patch, grep, glob, shell execution, web search, URL fetch, MCP server tools, local execution, Docker-backed execution, and package-building workflows.

Can Sunder Agent build a package?

Yes. Sunder Agent Builder lets Agent help create, edit, build, test, and iterate on Sunder package projects from inside Sunder App.

What is Sunder.Package.Agent.Contracts?

Sunder.Package.Agent.Contracts is a NuGet contracts package used by Agent extensions. It is not itself a Sunder runtime package.

Product proof

Agent is not a separate chatbot app.

Agent profiles, sessions, tools, package panels, and workspace context live inside Sunder App. The screenshot shows Agent installed as a Sunder package, alongside package panels and chat.

Sunder App showing the Sunder Agent package workspace with agent chat and package panels.

Package map

Choose the abilities you want installed.

Agent is a package family. The core handles profiles and sessions; separate packages add models, files, shell, web, MCP, memory, subagents, execution, and Builder.

Core

sunder.package.agent

Sunder Agent

Adds AI chat sessions, profiles, permissions, and tool orchestration inside Sunder.

Model providers

Provider

sunder.package.agent.provider.openai

OpenAI Provider

Lets Agent use OpenAI chat, embeddings, streaming, images, PDFs, and tool calls.

Provider

sunder.package.agent.provider.anthropic

Anthropic Provider

Lets Agent use Claude models with streaming, images, PDFs, and tool use.

Provider

sunder.package.agent.provider.gemini

Gemini Provider

Lets Agent use Gemini chat, embeddings, and multimodal input.

Provider

sunder.package.agent.provider.lmstudio

LM Studio Provider

Connects Agent to local models served by LM Studio.

Workspace tools

Tool

sunder.package.agent.tools.files

Workspace Files

Lets Agent read, search, create, edit, and patch selected workspace files.

Tool

sunder.package.agent.tools.shell

Workspace Shell

Lets Agent run approved non-interactive commands through a selected execution target.

Tool

sunder.package.agent.tools.web

Web Tools

Lets Agent search the web and fetch pages as markdown, text, or HTML.

Workflow extensions

Workflow

sunder.package.agent.mcp

MCP Integration

Connects Agent profiles to tools exposed by local or remote MCP servers.

Workflow

sunder.package.agent.skills

Skills

Adds reusable instructions and workflows from local paths or GitHub URLs.

Workflow

sunder.package.agent.subagents

Subagents

Lets a main agent delegate focused work to specialist child agents.

Workflow

sunder.package.agent.memory.semantic

Semantic Memory

Helps Agent recall relevant notes, preferences, project facts, and standing instructions.

Execution targets

Execution

sunder.package.agent.execution.local

Local Execution

Runs file and shell tools on the local machine when you allow it.

Execution

sunder.package.agent.execution.docker

Docker Execution

Runs workspace tasks inside Docker-backed execution targets.

Development controls

Development

sunder.package.agent.builder

Sunder Agent Builder

Lets Agent help create, edit, build, and test Sunder packages from inside Sunder.

Model providers

Choose where the intelligence comes from.

Use hosted models, local models, or a mix depending on the task and your setup.

OpenAI

Use OpenAI models for chat, embeddings, streaming, image/PDF input, and tool calling.

Anthropic

Use Claude models for chat, streaming, image/PDF input, and tool use.

Gemini

Use Gemini for chat, embeddings, and image, PDF, audio, or video input.

LM Studio

Use local models through LM Studio and discover available local models from Sunder.

Use cases

Use Agent for work that needs more than a text box.

Sunder Agent is strongest when the model can use installed tools: files, commands, web pages, MCP servers, local models, memory, subagents, execution targets, and Builder.

Ask questions about a selected project, folder, package, or local workflow.
Fix code, update files, run builds, inspect errors, and iterate through a task.
Create a dashboard package for a service, game server, 3D printer, smart home, or business workflow.
Turn scripts and commands into safer repeatable Agent workflows or package UI.
Research docs on the web, fetch pages, and apply the findings to local files.
Connect MCP servers to expose external services and specialized tools.
Use local models for private or offline-friendly experiments through LM Studio.
Let Agent Builder manage package development sessions for new Sunder packages.

Safety model

Local access is added intentionally.

Agent does not need every local capability by default. You decide which provider, file, shell, web, MCP, execution, memory, subagent, or Builder surfaces are installed and configured.

File access, shell access, web access, MCP, and execution are separate packages you choose to install.
Agent uses configured tool surfaces instead of silently receiving every possible local capability.
Provider secrets are stored through Sunder package configuration and secrets surfaces.
Local execution and Docker execution are separate choices, not mandatory defaults.

Start modular

Install Sunder App, add Agent, then give it the tools you need.

Install from the Sunder Registry or use the CLI.

Download Sunder Install Agent Join Discord

Backed by

Agent questions, answered directly.

Is Sunder Agent a separate app?

No. Sunder Agent is a package family installed into Sunder App.

Do I need every Agent extension?

No. Install the Agent core, then add only the providers and tools you want: files, shell, web, MCP, memory, subagents, local execution, Docker execution, or Builder.

Can Sunder Agent build packages?

Yes. With Sunder Agent Builder, Agent can help create, edit, build, and test Sunder package projects from inside Sunder App.

Can Sunder Agent use local models?

Yes. The LM Studio provider connects to a local OpenAI-compatible endpoint and lists local models.

Does Sunder Agent support MCP?

Yes. The MCP extension supports configured MCP servers and exposes MCP tools through Agent profiles.

Can Sunder Agent access my files or shell?

Only through installed tool packages and configured access. File tools, shell tools, web tools, MCP, and execution targets are separate capabilities you choose to add.