# Settings
## Settings
\
The **Settings** tab is where you define, refine, and fine-tune how your agent behaves—both as a trader and as an AI personality.
***
### Agent Status & Early Access
At the top of the page you’ll see the **Agent Status** card:
* Shows whether your agent is **Active** or **Deactivated**
* Displays your **CDX balance**
* Lets early supporters with **100,000 CDX on Base** claim **free early-access credits** to run the app
This is the operational control center for starting or pausing your agent.
***
### Core Personality
Shape how your agent communicates and responds:
* **Personality Description** – The fundamental character (e.g., professional, degen, humorous).
* **Communication Tone** – Formal, conversational, or anything in between.
* **Response Length** – Concise or more elaborate, depending on your needs.
Adjust these to give your trading agent a unique voice and presence.
***
### Thinking & Learning
Define the way your agent reasons and improves:
* **Direct** – Fast, simple tool calling. Best for straightforward tasks like checking a price or closing a position.
* **ReAct** – Think → Act → Observe loop. The agent reasons, takes an action, observes the result, then decides the next step. Great for multi-step tasks.
* **Chain of Thought** – Analyze → Plan → Execute. Sequential reasoning where each step builds on the last, producing a clear logical chain. Ideal for complex analytical tasks.
* **Graph of Thought** – Multi-perspective branching analysis. The agent explores the problem from multiple angles simultaneously, then synthesizes the best insights into a final answer. Best for nuanced decisions.
***
### Trading Approach
This is where you control the **guardrails for every trade**.\
The trading style and strategy defined during agent creation are stored here for easy editing:
* Update risk parameters and execution logic
* Select your preferred exchange
* Refine trigger commands
* Enforce TPs and SLs
* Adjust growth targets and trading philosophy
> **Important:** If you want to modify your agent’s trading strategy later, this is the place to do it.
***
### Optimizations
Cod3x lets you fine-tune how each goal uses compute resources.\
These **Credit Usage Optimization** settings help reduce costs without affecting overall performance.
* **Skip Final Chat Formatting**\
Returns raw model output instead of formatted text. much simpler output.
* **Skip Step Completion Assessment**\
Bypasses additional verification logic after each step. This slightly increases speed and lowers credit use at the cost of skipping LLM self-checks.
* **Enable LLM Model Auto-Selection**\
Allows Cod3x to automatically choose the most efficient model for each task based on complexity — using smaller models for simple steps and larger ones when needed.
* **Enable AI Commit Messages**\
Allows AI to generate commit messages when making code changes, improving consistency accross versions.
***
### Advanced Parameters
The **Advanced Parameters** section lets you fine-tune how your selected model behaves when running a goal.\
Cod3x supports hundreds of LLMs across providers like Anthropic, OpenAI, and xAI — each optimized for different types of reasoning and execution.
If you have **Enable LLM Model Auto-Selection** in the **"Optimizations"** modal, then you can select each LLM model for the 3 main reasoning steps of Cod3x' workflow.
**Claude Haiku 4.5** as the Primary and Simple models and **Sonnet 4.5** as the Main Analysis model is a good balance between cost and good results. For optimal results use **Sonnet 4.5** on all three spots.\
Our internal testing shows that Sonnet offers the best balance between **analytical precision**, **execution stability**, and **cost efficiency**.\
However, users can freely switch to alternative models depending on preference, cost targets, or desired response style.
Our research has shown that **Haiku 4.5** + **Kimi** is a very cost effective solution with solid results as well.
***
**Model Settings**
* **Temperature** – Controls output creativity.\
Lower values produce focused, deterministic responses; higher values encourage more exploratory ideas.
* **Top P** – Adjusts the probability range of word sampling, influencing output diversity and tone.
* **Frequency Penalty** – Reduces repetition across responses, ensuring less echoing or redundancy.
* **Presence Penalty** – Encourages the model to explore new topics and avoid looping around previous ideas.
These parameters allow advanced users to shape model behavior for **consistent execution**, **custom style**, or **maximized reasoning depth** — depending on the task.
> ⚙️ *Tip:* If you’re unsure where to start, use the default settings. Cod3x automatically optimizes these values for balanced performance.
{% hint style="info" %}
These settings won't be live during the beta version.
{% endhint %}
***
### Feature Availability
Many personality and reasoning features are still being finalized.\
Except for **Trading Approach**, which is fully live, some options (like deep adaptive learning or full social engagement) may remain partially functional during beta.
