Code & development
Agents that work through programming tasks, produce changes, and inspect what they’ve made.
From a request to a working change.Agents need
room to work.
A model can decide what to do. Hatch is the sandbox for the work that follows.
Discuss early accessA task for your agent
“Turn this dataset into a report.”
HatchAgent workspace
dataset.csv
SourceValue
report.md
A useful result.
An agent. A workspace. A result.
Illustration of the sandbox conceptThe model is only
part of the system.
Agents turn a conversation into a sequence of actions. Those actions need somewhere to happen. That’s the role of a sandbox.
From request to resultExample workflow
Request
“Take this dataset and prepare a report.”
Agent
Break the task into steps.
Workspace
Work with the inputs.
Carry out the steps.
Inspect the output.
Result
A report the application can use.
A place between
intention and output.
The model supplies direction. The surrounding application supplies the task. A sandbox provides an environment in which that task can be carried out.
We’re building Hatch around this part of the agent stack: the workspace.
Tell us about your agentWhat will your
agents work on?
These are the kinds of workflows we’re exploring with builders. Bring us yours.
Data & research
Agents that work with supplied information and turn it into analysis, reports, or other useful artifacts.
From raw inputs to a useful result.Multi-step workflows
Agents that need to carry context through a sequence of actions instead of returning a single answer.
From one instruction to several actions.A few things to know.
What is Hatch?
Hatch is the working name for Alabric’s AI agent sandbox offering. It focuses on the environment where an agent carries out its tasks.
Can I start using it today?
We’re discussing early access with builders. Tell us about your use case so we can discuss fit and availability directly.
Which tools and runtimes will it support?
The launch scope is still being defined. Share your runtime, tool, connectivity, and persistence requirements with us; the illustrations on this page are concepts, not a specification of supported features.
Bring the agent.
Let’s discuss the environment.
Tell us what it needs to do, and what’s getting in the way.