Agent systems / Research preview

Agent Laboratory

A native Mac workspace for people building AI agents. Inspect runs and their evidence, trace decisions, revise memory with its history intact, and prepare controlled learning experiments.

Agent systems

See what your agents are learning.

An agent's final answer rarely explains everything that happened along the way. Agent Laboratory is a native Mac research workspace for examining runs, evidence, decisions, and memory, so developing an agent becomes a more inspectable experiment.

Made for

For people building AI agents who need to understand behavior before changing it, compare an idea against its evidence, and keep a useful history of what they tried.

Try it / follow along

Research preview

We’re still building. Get in touch if you’d like to collaborate or hear more.

Talk about Agent Laboratory

A closer look / the workflow

From the first step
to something useful.

Start with the run
Inspect what an agent did and the evidence attached to the work. Keep the observed behavior separate from the explanation you hope will account for it.
Trace a decision
Follow decisions and memory through the workspace, looking for the point where an assumption, instruction, or earlier experience shaped the result.
Prepare the next experiment
Revise memory with its history intact and set up a controlled learning experiment around a question you can actually investigate.

Inside Agent Laboratory

The details
make the difference.

Evidence beside behavior

Run inspection keeps concrete traces near the agent's output, giving a researcher more to work with than a convincing summary alone.

Memory with a history

Revisions preserve context about change, helping distinguish a deliberate update from an unexplained difference in later behavior.

A Mac research workspace

The native application brings these activities into one environment, making it easier to move between inspection and the next research question.

Where it stands

Research preview

Agent Laboratory is a research preview focused on run inspection, memory history, and experiment preparation. Get in touch to explore how the workspace could fit your agent research.

Ask about Agent Laboratory