Reviewing Performance and Debugging in Real Time

Cod3x agents are designed to learn, adapt, and improve with every run, but that doesn’t mean they’ll always get things right on the first try. Sometimes a trading strategy underperforms. Sometimes a task doesn’t execute the way you expected. That’s where Cod3x’s debugging tools and more importantly, the conversational interface step in.

If something feels off, you can now ask your agent directly what happened, just like you’re chatting with a teammate. You might say:

“Why didn’t you post yesterday?”

“Today’s PnL is awful, what went wrong?”

“Please update the strategy to avoid these kinds of losses.”

The agent won’t just explain what happened, it will take action. Cod3x can analyze your past runs, identify points of failure or misalignment, and update your tasks, strategies, or tone based on the conversation. That means you’re not stuck with static settings or forced to manually reconfigure things. Whether you’re frustrated by poor trading results or just curious about what happened during a specific task, your agent can walk you through the logic step by step and suggest meaningful changes in response.

This conversational loop is now the primary way to manage and improve your agent’s behaviour. But for those who want a deeper, more structured look under the hood, the Runs Interface is still available and powerful.

Within the Advanced dashboard, under the Scheduler tab, every agent task (formerly called “goals”) generates a history of executions, giving you full transparency into what the agent tried to do, which tools it used, and how it performed. Clicking on a past run reveals a complete timeline of execution: the original instructions, the tools called, the logic followed, and the outcome. You’ll see whether each step succeeded or failed, and if something went wrong, you’ll get a breakdown of exactly why, whether it was a missing tool input, a failed API call, or a logic mismatch in the prompt execution.

At the bottom of each expanded run, Cod3x provides intelligent recommendations based on what went wrong. Maybe the trading logic needs to be more conservative, or a tool should be swapped for something more reliable. These suggestions can be accepted right away and will immediately update the relevant strategies or tasks without any manual reconfiguration. It’s debugging built into the workflow, not a separate process.

If you want to go even deeper, raw execution logs are available at the bottom of each run. These show every interaction and decision the agent made during the task, down to plugin-level details and internal success criteria. This level of granularity is useful for power users who want to fine-tune performance or understand why two similar runs led to different outcomes.

And just like in the conversational flow, these insights aren’t just for reading. You can take action instantly. Update your trading strategy. Rebalance your tone. Change how tools are used or how tasks are triggered. It’s all editable on the spot, directly from the agent’s interface.

So whether you’re debugging like a power user or simply chatting with your agent to understand what went wrong, Cod3x gives you full visibility and the power to adapt.