What are Truffle apps?
Getting started with an agentic toolkit
A Truffle app is a set of tools you give to your Agent to accomplish a task. Think of it like an MCP server that lives in an ephemeral, trusted container.
At it’s core, it turns a python script into a box of tools.
Where does this stuff run?
By default, when using the client - you are connecting to an instance of the truffle firmware in the cloud. You can switch to hardware if you place an order for a Truffle-1 at anytime ;)
The cloud is a secure environment that spawns siloed containers for apps. Your app is running in a PC in our datacenter, so does not have access to your local files unless you attach them
Available Models
- Deepseek-R1
- Qwen-3-235B
| Advantages | Disadvantages |
|---|---|
|
|
The cloud is a secure environment that spawns siloed containers for apps. Your app is running in a PC in our datacenter, so does not have access to your local files unless you attach them
Available Models
- Deepseek-R1
- Qwen-3-235B
| Advantages | Disadvantages |
|---|---|
|
|
Runs locally on your hardware, providing full privacy and unlimited inference time.
Available Models
- Qwen-3-30B
- Qwen-2.5-32B
- Deepseek-R1-Distill-14B
| Advantages | Disadvantages |
|---|---|
|
|
How apps work
Truffle apps are toolboxes and logic available to your model in a continous loop. ChatGPT style systems are essentially agents with a single tool; respond_to_user.
Models using our runtime do not by default respond to their user, but instead loop over tools until they complete a short-medium term task as defined by the user prompt.
ChatGPT, for example, is simply an agent, with one tool. With truffle, you can give your agent arbitrarily many capabilities.
App Selector
As the image suggests, the model lives by itself ready to be slot into a sandbox of tools. These sandboxes are what we call apps
Terminology we’ll use
| Term | Description |
|---|---|
| App | A set of tools given to a model to pick from, when executing a task within a runtime |
| Tool | A function that returns context to update the state of the model |
| Tool Rollout | A sequence of tool calls where each subsequent tool call depends on all previous tool calls |
| Task | An initial prompt, that allows the model to perform a tool rollout |
Features
Out of the box, the SDK today has a few features that set it apart from other agent frameworks.
The simple ones are stuff like;
- Guaranteed tool execution (no hallucinations)
- Custom prompts
- Conditionals
- Async tools
- Context management
- Networking
Click on a card below to explore these features
Conditionals & Predicates
Use boolean logic to expose tools to your models
Async Tools
Integrate asynchronous operations with ease, from I/O to external APIs.
Context Usage & Management
Maintain and manipulate execution context across complex tool rollouts.
Networking with Truffle-1
Connect multiple Truffle-1s for distributed workloads and agent coordination