A space of its own.
A dedicated copy to work in. The original project stays outside the worker.
OPEN SOURCE. LOCAL FIRST.
Your agents
move forward.
You stay
in control.
A dedicated workspace for your AI agents.
Defined access. Traceable actions.
Apache-2.0 · In development
ISOLATE. CONTROL. OBSERVE.The project
Delegate the work.
Keep the boundaries.
Giving an agent a task should not mean handing it the keys to your entire machine.
ai-guard organizes execution around a working copy, explicit decisions and a log independent of the worker. You review the results before choosing what to keep.
Clear rules.
A decision before launch, runtime restrictions and visible denials.
A readable trail.
See what was requested, allowed and observed. Identify what remains unknown, too.
Move forward.
Then review.
Assign a coding task, run a command or review an agent’s work: the same workflow.
The local dashboard described in the project brings together actions, decisions and events. It helps you understand execution; control stays with the broker and runtime.
- 01
Prepare.
Create a filtered copy of the project for the session.
- 02
Run.
Submit an invocation to the broker under the selected profile.
- 03
Review.
Inspect the available events and results.
- 04
Export.
Export files to a new destination.
The code is open.
So is what comes next.
Read. Try. Contribute.
Explore the repository, test the project against your needs and help shape its future. Local mode is designed to work without a mandatory cloud account; connected agents retain their own requirements.
Explore on GitHubIn development. Production readiness still needs to be established for each environment. Cloud support is a project goal.