2025 Spark Update: Spark AI Community Support, Spark Chat (Gemini), and Multi‑Agent Knowledge

Spark is one of the most popular ways to turn community knowledge into instant AI answers. In the 2025 Spark update, Spark expands community support with a dedicated support AI bot, better knowledge handling, and a new Spark Chat experience built on Gemini.

Below is a clear breakdown of what the update changes and how you can configure it—so your community can get help even when moderators aren’t available 24/7.

What the 2025 Spark update does for community support

The core update introduces and strengthens Spark as an AI community support layer. Spark can automatically answer member questions by using your configured community knowledge.

The speaker describes Spark as a support AI bot that helps members get instant answers even when they’re not on Discord all the time. The update also positions Spark for two common needs:
- Smaller projects that want immediate, automated support
- Larger communities that want to reduce repetitive moderator responses

Overall, the 2025 Spark update is focused on making Spark easier to test and more effective at answering questions grounded in your own knowledge.

How to test Spark by adding the bot and tagging it

If you want to get a feel for how Spark behaves in your community, the update emphasizes a straightforward testing approach:
1. Introduce the Spark bot into your community.
2. Tag the bot so it can respond—helping you observe how Spark answers.

This quick start is meant to let you evaluate Spark’s behavior before you invest time into more advanced configuration.

Custom knowledge: integrations and manual document uploads

A major part of the update is how you provide knowledge to Spark so it can answer based on what your community already knows.

The update describes custom knowledge setup as having two main components:
1. Adding knowledge (including knowledge integrations)
2. Creating a multi-agent setup (covered in the next section)

Knowledge integrations

For knowledge inputs, the update includes integrations that help connect external knowledge sources into Spark. The transcript specifically calls out an integration with Gbook.

Manual text uploads

In addition to integrations, Spark also supports manual uploading of text documents. The speaker mentions that you can upload text you already have, including content like:
- GitHub-related information
- Technical documents

The goal is to let Spark use your configured knowledge so answers are tied to your community’s materials.

New multi-agent setup for different roles across channels

The 2025 Spark update introduces multi-agent support. Instead of limiting one channel to a single purpose, you can set up multiple agent roles and use them across multiple channels.

In other words, different agents can specialize in different kinds of community interaction. The transcript describes the concept at a high level using roles such as:
- A welcoming/jokes agent
- A technical support agent

This is intended to make community responses feel more organized and relevant, while still leveraging Spark’s AI-driven answers.

Backend improvements: smaller chunking to reduce hallucinations

Under the hood, the update improves how Spark searches through your documents and generates answers.

A key backend change is that Spark now chunks documents into smaller sizes. The transcript explicitly links this to two outcomes:
- Better search behavior
- Less hallucination

The speaker frames this as important because the way documents are divided affects how the system retrieves relevant information during answering.

Gemini model migration and Spark Chat conversation improvements

Beyond community support, the 2025 update also includes upgrades to Spark Chat.

Model migration from OpenAI to Gemini

The transcript states that Spark’s model migrated from OpenAI to Gemini. The reason given is to better align with the core team’s speaking style.

Spark Chat: expression, style, and longer “contact lenses”

The Spark Chat update highlights Gemini’s strengths in conversation—particularly:
- Expression
- Matching the talking style you want

The transcript also introduces the idea of “contact lenses” being optimized for longer conversations. In this context, “longer contact lenses” means the system can use more context so the conversation feels more personal.

The speaker notes that with longer contact lenses:
- The conversation feels more personal
- The system remembers more about past conversations

To try Spark Chat, the transcript suggests installing the bot and clicking Spark Chat on the website.

Conclusion

The 2025 Spark update expands what Spark AI can do for community support. You get a support AI bot for instant member answers, easier initial testing by adding and tagging the bot, improved custom knowledge setup (including integrations like Gbook and manual text uploads), and a new multi-agent setup for role-based responses across channels.

On the backend, Spark improves document chunking to support better search and reduce hallucinations, and it migrates to Gemini. Finally, Spark Chat focuses on Gemini-powered conversation quality, emphasizing expression and style, with longer context (“contact lenses”) designed to make interactions feel more personal.