dosus-new-fact-based-reasoning-agent-is-generally-available

In our May Dosu Drop blog we teased our new Dosu agent, but today we're excited to announce that we have rolled out this experience to everyone. We've been floored by the reaction from folks who were part of our early staged rollout, and that's a big part of why today we're confident we can make this generally available to everyone.

One user with early access went as far as saying this is the "first time AI has genuinely impressed me."

A Smarter, Faster, Self-Improving System#

We've re-engineered the Dosu agent to center around fact-based reasoning. Now, when Dosu needs to solve a problem, it starts by taking stock of what it already knows. If it has all the information it needs, it can respond immediately. If it doesn't, Dosu does research to find missing facts and remembers them for next time.

As the knowledge base grows, so does its speed and performance.

Our engineers have been nerding out about this new design, but more importantly the reaction from the community has been overwhelmingly positive.

Here are a few reactions to the new Agent on GitHub:

This new agent architecture also allows for:

Big-picture understanding: By using a knowledge-base structured around topics and concepts, Dosu should perform better when answering questions that require higher-level knowledge that is distributed across many different sources.
Better steering: Dosu is now more attentive to user-provided descriptions of data resources and channels, offering more controls for users to customize their ideal response behavior.
Improved introspection: You can clearly see what Dosu's goals were, what assumptions it made, and how it interpreted the sources it's citing. This visibility means more opportunities to understand the behavior of the Agent and to give actionable feedback when it makes mistakes.

Agent Objectives: More Controls and Customization#

Dosu now lets you customize and standardize reply behavior across all platforms. Want Dosu to only focus on de-duplicating requests in your github repo? You got it. Want Dosu to only surface relevant documentation? Absolutely. Want Dosu to suggest workarounds and code modifications? No problem.

You can also now customize Dosu's behavior using natural language, right from the dashboard. This is similar to how Cursor users add .cursorrules or agent.md files to control agent behavior. Examples include telling Dosu to always reference certain parts of the codebase for certain type of requests, tips and tricks for how to correctly identify duplicate requests, or how to reason about the changelog files.

What's Next#

We're working towards the promise of being your trusted knowledge generation and maintenance partner, freeing you up to focus on building. We have more announcements coming next week for our June Dosu Drop, so please be on the lookout for that.