Photo by

Navigating AI Solutions: The Balance between Creativity and Business Accuracy

How I Created an assistant for Character.AI that actually works.

Sep 22, 2023
3 min read


The world of artificial intelligence is as diverse as it is revolutionary. Tools and platforms pop up, each promising a unique flavor of AI magic. Today, we're focusing on two distinct players in this realm: Character.AI and OneAgent. At first glance, they might seem worlds apart — one's all about spinning tales, and the other's laser-focused on clear-cut data. So, why put them on the same stage?

It's all about the broader perspective. By examining these two side-by-side, we not only appreciate the individual strengths of each platform but also highlight a fundamental truth about AI: it's not one-size-fits-all. The spectrum of AI encompasses both the heart (creativity, emotion, narrative) and the head (accuracy, reliability, logic). In navigating this terrain, it's invaluable to understand the distinctions and where each tool shines. By diving deep into the philosophy of Character.AI and OneAgent, we can better grasp the multifaceted nature of artificial intelligence and recognize that every tool has its moment, its purpose, and its audience. 

In this exploration, we'll delve into the core of Character.AI, understanding its role and strengths, and similarly, unpack the essence of OneAgent and its key functions. Following that, we'll take a hands-on approach: constructing an AI Agent using the OneAgent tool, populating it with information from the Character.AI website, and then putting it head-to-head with the Assistant powered by Character.AI itself. So, let's get going.

For those too pressed for time, here's a quick glimpse of our comparison: assistant I built using OneAgent’s official assistant
Result: accurate + fact check Result: Misleading

Character.AI: A Deep Dive into Conversational Creativity

Born from the minds of AI experts Noam Shazeer and Daniel De Freitas in 2021, Character.AI emerged as a trailblazer in the conversational AI domain. The platform offers users the unique capability to craft a fully customized AI companion, each characterized by its own distinct personality and values. But what lies at the heart of this platform?

The Heartbeat: C1.2 Model

At the heart of Character.AI is its C1.2 model, a refined iteration of its predecessor, C1.1. While C1.1 set the stage with its focus on entertainment, roleplay, and fostering emotional connections, C1.2 takes it a notch higher. With this new model, users don't just engage in casual chit-chat; they can harness its capabilities for practical tasks like drafting emails, brainstorming ideas, and test preparation. As the model is currently in its early stages, Character.AI is deploying C1.2 to select characters, with plans of broader rollout as they fine-tune accuracy and personality nuances.

Strengths: The Art of Digital Storytelling

At the core of Character.AI's capabilities lies its knack for digital storytelling. This is where the platform truly stands out. Here are its primary strengths:

Easy Start: To engage, all users have to do is choose a character. Want something more personalized? Creating your own character is straightforward. Simply follow their comprehensive guidebook, which effectively breaks down the process into filling out basic fields like the Name, Greeting, and Description fields. For those who want to add some spice, there’s a Definition field to add more nuance. Once that's done, the AI takes over, bringing your character to life.

Interactive Dialogue Creation: Character.AI's model is adept at crafting dialogues that are not just responses but interactive narratives. Conversations with this AI feel alive, filled with unexpected twists and turns that captivate users.

Personalized Interaction: The AI tailors its dialogues to suit individual user preferences, ensuring each interaction is unique. The personal touch is evident, making the experience immersive.

Emotional Connections: Unlike many AI platforms that can feel robotic, Character.AI builds an emotional connection with users. It understands sentiments and crafts responses that resonate on an emotional level.

Versatile Use Cases: Character.AI isn't just a tool for entertainment. Its capabilities stretch beyond that, making it apt for educational purposes, helping users with introspection, or even as a companion for those moments of reflection. The potential is vast, and the possibilities are endless.

Limitations: Another side of the coin

While Character.AI is a master of imaginative dialogues, its strength can sometimes be its Achilles' heel. Let's look at the limitations:

Hallucinations: The very trait that enables the AI to create imaginative tales can also lead it to provide information that, while fascinating, might not be factual. This is termed as 'hallucinations' in the AI world.

Character inconsistencies: Due to the limited amount of conversation context that a character can consider, they might appear forgetful, not recalling details shared earlier in the conversation.

This results in scenarios where characters might claim knowledge or experiences that conflict with their established background or what's been shared in the dialogue.

Limited Utility for Factual Queries: If you're seeking precise, fact-based answers, especially on technical or specific topics, the platform might occasionally weave in imaginative elements, blurring the line between fact and fiction.

Reliability in Specific Domains: For queries related to its own functionalities or guides, the AI might sometimes falter, leading to answers that are more whimsical than accurate.

I asked Character.AI’s Assistant about the availability of an API for integration into other apps, and it confirmed they had one. However, upon checking their FAQ section, I found that they currently don't offer an API. They did suggest reaching out via email for more information. Here's the result screenshot:'s Assistant "Is there an API answer"

You can also notice the previous answer, where Assistant says that their model is partially based on GPT models, which is also not true.

In another example, where I asked what is the character limit for the definition field, which is 3200 characters and there’s no possibility to add more Definition fields. Here’s what Assistant says:'s Assistant "Definition field limit" answer

This highlights the occasional inconsistency in responses and underscores the importance of verifying crucial details from more direct sources.

Introduction to OneAI's OneAgent

What is OneAgent?

OneAgent is more than just an AI tool; it’s your personalized AI Agent crafted to answer questions and execute tasks based on the specific content you feed it. Think of it as a truly transparent, explainable agent. When questions stray off-topic or venture outside its trained domain, OneAgent gracefully declines, maintaining its role as a reliable, on-topic assistant.

How Does It Work?

OneAgent operates on a principle of strict adherence to the content you've entrusted it with. This could be in varied formats: from web links, PDFs, and knowledge bases to more dynamic content like YouTube videos, podcasts, or data from internal systems like CRMs. In essence, OneAgent acts as your personal "walled GPT", with the unique ability to ingest vast amounts of diverse content, irrespective of size.

Here's the magic: Once OneAgent absorbs all this data, it becomes a powerhouse of precise information. Pose a question, and it sifts through the ocean of content you’ve provided, pinpointing the exact piece of information that answers your query. And for added transparency, every answer comes with a direct link or reference to the source, so you always know where the information originated.

What sets OneAgent apart is its blend of out-of-the-box functionality with customization. Right from the start, it's set to provide reliable answers. But if you wish to tweak its behavior or set specific responses, OneAgent is flexible, allowing for personalization that suits your unique requirements.

Strengths of OneAgent

Works out of the box: OneAgent works efficiently right from the get-go. But its design allows for continuous learning and adaptation, meaning it becomes even more efficient as you provide it with more content and fine-tune its responses.

Transparency and Trustworthiness: One of the prime strengths of OneAgent is its unparalleled transparency. Every answer or piece of information provided by the agent directly traces back to the source. This not only instills trust but also eliminates the ambiguity often associated with AI responses.

Customizability: Users can customize Agent’s responses and behavior to better align with specific needs, ensuring that the AI assistant truly feels like a tailored solution for each user.

Versatility in Content Absorption: OneAgent can ingest a diverse range of content formats. Whether it's a podcast, a PDF, a YouTube video, or data from internal systems, this AI agent is adept at sifting through various types of data sources to fetch precise information.

Strict Adherence to Provided Content: Instead of wandering into the vast realm of the internet, OneAgent strictly sticks to the content it's been fed. This ensures the reliability of the answers and minimizes chances of inaccuracies or deviations.

Efficient Data Management: The system is designed to handle vast amounts of content, ensuring that irrespective of the size or complexity of the data, OneAgent remains efficient and swift in its responses.

Limitations of OneAgent

Dependent on Input Quality: OneAgent is only as good as the content it's fed. If the information provided to it is outdated, incorrect, or ambiguous, the agent will still use it as a reference, leading to potential misinformation.

Lacks Creative Generation: Unlike platforms that emphasize creativity, OneAgent is strictly factual and doesn't generate new or imaginative content. This makes it less suitable for roles requiring creative dialogues or narratives.

No Broad Knowledge Base: While many AI systems boast a vast general knowledge base, OneAgent strictly adheres to the content it's provided. This ensures that there are no "hallucinations" or fabrications. However, it also means OneAgent won't have answers to general queries outside its trained domain. This focused approach is a trade-off for ensuring accurate and consistent information based directly on the content ingested.

Requires Regular Updates: As it relies on the data ingested, OneAgent needs regular content updates to stay current. It won't automatically know the latest trends or news unless it's updated with that information.(automated updates available only in paid version)

Strictly Professional: Its design leans more towards business and professional uses. OneAgent might not be the best fit for casual or entertainment-oriented applications due to its strict adherence to facts and data.

How I build an Agent based on’s database

Step 1: Name Your Agent & Plug in the Link

First things first, I named my agent “Character AI” - seemed fitting. 😀

I then popped in the link to's guidelines, their FAQ section, and a few select articles from their blog. There's an option to allow the Agent to crawl within the domain or even venture outside it, plus you can set the depth of the crawl. 

Here's another screenshot.

Step 2: Start Conversing

Honestly, that's it! No tedious setup, no extensive configurations. The system instantly processes the data and is ready to engage in a conversation. They even throw in some suggested questions to kick things off. Remember when we quizzed's Assistant about the API? Let's click on that question:

And the answer is:

Spot on, clear, and concise! Oh, and check out that Fact Check button. It lets you trace the source of an answer. Let's give it a whirl:

And there you have it, the source laid out right in front of us. In our case, it's a single source, and we can see the exact quote it pulled from the FAQ section.

Alright, let's continue asking our Agent the same questions to ensure consistency and verify the accuracy of its responses based on the ingested content. This will also give us insights into how well OneAgent retains and recalls specific details, allowing for a more informed comparison with's Assistant.

Next questions:

1) “Tell me about the Character AI's model.” 

OneAgent (top) vs (bottom)

2) "What's the maximum character count for the definition field in"

OneAgent (top) vs (bottom)

After demonstrating OneAgent's precision, it's also worth highlighting its adherence to the information provided. In this next screenshot, I posed a question outside of its trained domain. As you'll see, OneAgent doesn't venture a guess or provide generic answers; it stays true to its design by declining to answer anything it hasn't been specifically trained on. It's a reminder that while its precision is laudable, its scope is strictly defined by the content it's been fed.

Free topic answer: OneAgent (top) vs (bottom)

Seeing is believing! After our comparison, it's clear how effective OneAgent can be. Try it for yourself here.

Comparison table: vs OneAgent

Parameters OneAgent's Assistant
Purpose/Focus Business-centric, answers and tasks based on ingested content. Primarily storytelling, character interaction, and narrative.
Ease of Setup Simple: Name your agent, input the content source, and start conversing. Quite straightforward: choose or create a character by filling out basic fields and let the AI take over.
Data Source Digests content in various formats: HTML, videos, PDFs, podcasts or internal systems. Trained on a wide variety of text data, including books, articles, blogs, forums, social media, and many more sources.
Response Precision Strictly user-provided content; precise answers directly sourced from this content. Uses a vast knowledge base to create expansive, interactive dialogues.
Interactive Experience Concise and directly sourced, ensuring accuracy. Answers can be creative, expansive, and not always precise.
Interactive Experience Straightforward Q&A format, precise but less interactive in a narrative sense. Offers dynamic, emotionally-resonant dialogues with unexpected twists.
Business Readiness Business-ready, primed for business use, especially in areas like customer support or accessing a knowledge base. More suited for entertainment, though could be used in business scenarios requiring interactive narratives.
Educational Potential Best suited for precise QA style education, where accurate answers are needed from provided content. Can be used for content generation: writing essays, dialogues, campaigns, etc. Be aware of hallucination risk.
Transparency Directly shows the origin/source of its answers for full transparency. Source of answers is the AI's training and algorithms, without direct attribution to specific content.
Limitations No broad knowledge base. Answers only from user-provided content, so can't handle questions outside of its trained domain. Creative expansiveness can lead to inaccuracies. Limited conversational context memory.


There it is, our side-by-side look at Character.AI and OneAgent. At the heart of it, it's a play between creativity and precision. While Character.AI shines with its interactive tales and emotional touch, OneAgent stands out as the no-nonsense, fact-first champ. It's a testament to the breadth of AI: whether you're seeking imaginative engagement or laser-focused information, there's a tool tailored for you. Choose wisely based on your needs!

Read Next

AI Expert Onboarding Session

Start smart - schedule a free session with one of our AI experts to set up and configure your agent for success