Hey HN, I'm Dennis and I'm the founder of Unblocked.

Unblocked provides helpful and accurate answers to any question about your codebase.

When a project and team gets big enough, one of the most time consuming parts of building software is understanding the entirety of your codebase — how it works, why it was written, and works the way it does.

Based on our collective painful experiences of countless hours in meetings, dealing with interruptions and digging through obsolete documentation while building software at companies like Microsoft, Amazon, facebook/Meta and Apple… we created Unblocked.

Unblocked provides helpful and accurate answers to any question about your codebase.

Our goal is to help you get the information you need so you can maximize the time you spend writing code.

We (securely) connect to the data sources teams use to talk about their codebase - source code providers, bug trackers, documentation/wiki systems and messaging platforms and expose the information we think might be helpful to you in your IDE, our dashboard and our macOS app.

Unblocked is a lot more than a wrapper around GPT4. Here's how we collect and analyze your provided data:

Document-type specific chunking techniques: Source code doesn’t chunk particularly well at the line-level. For example, an LLM needs surrounding context to understand relationships between classes and their methods to answer higher level questions about code.

Local context: Early generative apps like ChatGPT have no concept of what you’re working on right now. Unblocked helps short-circuit this process by leveraging our tight IDE integration to infer local context.

Reference generation: We’ve found that citing reference is important to help customers gain confidence in Unblocked’s capabilities. Using a handful of in-prompt and output-chain techniques, Unblocked is able to accurately reflect which reference documents were used by the LLM to answer questions.

On data privacy and security: No 3rd party models are trained on customer data For Unblocked’s own models, which we use for component classification tasks, each customer gets their own fine-tuned model, no sharing Authorization checks at network, API and data layers to enforce data isolation Per-team keys to encrypt secrets and data

I’d love for you to try out Unblocked. You can get started here: http://www.getunblocked.com/

Please give it a spin and let me know what you’d want to see improved!