Stable Concepts, Smarter Citations, and One-Click Model Swaps

In this week’s engineering demo, we walked through improvements to three key features: Concept Generation, Regenerate Answer, and Inline Citations. All are focused on building transparency, reliability, and better control over your workflows with Storytell.

Concept generation: clarity through structure

Shelby, our principal software engineer, gave us a preview of what’s coming in V3: a new way Storytell will identify and organize the key ideas inside your content. In this upcoming version, Storytell breaks down large amounts of data into a manageable set of themes so you can get the big picture without digging through every detail.

Concepts in V3 are structured to be durable. Once a seed—the smallest unit of meaning—is added to a Concept, it stays there. New seeds can be added as new content is ingested, but none are removed. This preserves the integrity of the Concept over time. If you're building actions or prompts around a Concept, you can count on its structure staying intact, no matter how much new data arrives.

Shelby emphasized how important this stability is, especially as Concepts become dependencies in workflows. Imagine creating a workflow around "Q1 Sales Performance." You want that anchor to stay consistent. If your needs change (e.g., different fiscal years), you can apply labels or spin up new Concepts entirely without affecting existing automations.

You’ll also be able to see a version history of each Concept. Think of it like a logbook: a timeline of the data that shaped it. Whether new information was added manually or evolved through the system, you’ll always have visibility.

A few takeaways:

• Concepts are additive only—nothing gets removed.
• You can create a new Concept if you want to redefine meaning.
• Version history and manual labels help preserve context.

Regenerate answer: testing across models made easier

Alex, our DevOps engineer, introduced a simple but powerful new tool: a regenerate button that lets you rerun the same prompt across different language models with just one click. Previously, if you wanted to compare outputs from Gemini, Claude, GPT, and others, you’d need to manually copy the prompt, switch models, and start over.

Now, the experience is much smoother. After generating an answer, you’ll see a dropdown button that allows you to select a different model. The prompt, context, and references stay intact, letting you compare responses without any extra effort.

This feature makes multi-model testing easier and faster, especially if you’re working on prompt design, instruction tuning, or trying to understand how different models interpret your data.

• Keeps prompt context and references intact
• Encourages side-by-side model comparison without losing flow

Inline citations: trust through traceability

Alex also shared improvements to inline citations, a long-requested feature aimed at helping users understand exactly where an answer comes from. When a response is generated by Storytell, it now includes small numbered markers (like [1], [2]) throughout the text. Clicking on one brings you straight to the original source, whether that’s a document snippet, spreadsheet row, or asset in your Collection.

This makes it easier to validate claims or trace key insights back to their origin. In his demo, Alex showed how the system attempts to place these citations at the end of sentences for readability. While the placement isn’t always perfect (since it depends on model behavior), we’re working on refining this with clearer system instructions.

With citations embedded directly into AI outputs, you can feel more confident reviewing, sharing, and refining the answers you get.

• Clickable markers directly link answers to data
• Improves transparency and user trust

The bigger picture

These features are part of our ongoing effort to build AI systems that support real workflows—not just generate text.

• Concept generation gives you dependable anchors in your datasets.
• Regenerate answer makes multi-model testing fast and frictionless.
• Inline citations make it easier to validate and explain AI-generated results.

Become an Alpha or Beta Tester

Get early access to features as we release them by becoming an alpha or beta tester. Here's how to sign up: https://web.storytell.ai/early-access