​

⏳ Time Travel

Time Travel lets you freeze time at any point in an agent run and replay that exact moment — as many times as you want.

Whether your agent failed halfway through a task or made an unexpected decision, you can zoom into that event, rerun it across multiple simulations, and debug it in isolation. Modify the model, adjust your prompt, or tweak instructions — no need to rerun the whole session or regenerate brittle, one-off states.

This is your tool for:

• Isolating failures without starting over and rerunning the whole session
• Debugging variability in model outputs
• Fixing bad decisions by adjusting prompts at critical steps
• Creating reproducible replays from fragile, real-world sessions

You’re not guessing anymore — you’re traveling back to the exact moment things broke and fixing it.

​

✍️ How to Create One

1. Select any Event
2. Click the Time Travel tab under the two timelines
3. When the screen loads, you will automatically be in a new Time Travel. We have froze time for right before your event i.e we have captured your exact LLM call and the state of your agent (with all the images). You can then modify any of this information and tweak it to debug your agent.
4. Write a name for your Time Travel
5. Choose a model and number of simulations (optionally adjust temperature and other parameters)

📷

6. Click Preview to see the input that will be passed to the LLM to make sure it looks right
7. Click Run Time Travel Simulations (bottom right) to run the Time Travel simulations!

📷

8. Optional: If you have images, you can select which images you want to use in the time travel simulation by checking the box next to the image name. This is useful if you want to test your input with and without vision.

​

🧮 What’s in the Output

​

How to access the results

• Click the Results tab to see the results of your Time Travel.
• Depening on how many simulations you ran, your connection speed, LLM model, etc., it may take a few minutes to finish, so sit tight!

​

Output style

• The outputs will include both the raw responses from the LLM and intelligently organized groupings of those responses.
• Each group features a descriptive summary of the agent’s typical actions and a clickable name to easily access the underlying raw responses.

The results are grouped by behavior category so you can instantly see dominant, edge, and problematic outcomes.

📷

​

⏳ Variables - Organize your prompt changes

Time Travel lets you modify any part of your agent’s state — but sometimes, you want to be systematic about it.

That’s where variables come in. Variables let you mark specific parts of your input as “unchangeable”, so you can experiment cleanly while keeping the rest of the input constant.

Use variables to:

• 🔁 Isolate changes in prompts, instructions, or environment
• 🔒 Lock critical context that shouldn’t change between runs

​

How it works

1. Click the Variables tab
2. Highlight a field in your input (e.g., a line of reasoning, fixed text, or an instruction).

📷

3. Scroll up and add a name for your variable — this tells us what you want to call it.
4. Click Set as Variable

📷

5. Scroll to the bottom of the page and you can now see all your defined variables

📷

6. Go back to the time travel page — Now your variables are fixed and don’t clutter your screen so you can focus on the parts of the prompt that you want to change.

📷

7. Optional: Check the exact input to your LLM by clicking the Preview button to ensure that everything still looks right!

📷

By clearly separating what you can and can’t change, you’ll get cleaner comparisons and sharper insights.

Variables are especially useful when debugging brittle prompts, testing fallback instructions, or analyzing model behavior across slight input changes.