GenAI has transformed nearly every aspect of software development, but one constant is here to stay: the need to understand code. Whether code is manually written by a developer or generated by AI, eventually, someone else will need to understand and interpret that code.
But utilizing AI to understand code requires more than a generic chatbot, it demands context that’s unique to your codebase. Think about it. When a developer on your team needs to understand a complex piece of code, there’s a 3am outage in part of your codebase that your on-call dev is not familiar with, or when orienting a new vendor; context that is unique to the nuances of your codebase is what distinguishes a generic answer from a truly useful one.
Today, we’re excited to announce the launch of /ask Swimm, an AI assistant that answers all of your developer’s coding questions using the unique context of your codebase.
Context is king
Some of the biggest names in tech are building AI code assistant tools. Think: GitHub Copilot, Google Duet, and Amazon CodeWhisperer. All of these tools are competing on their analysis and LLM capabilities, but at the end of the day, a tool doesn’t know what it doesn’t know.
Not only does /ask Swimm understand your codebase, but it also understands the logic that isn’t evident in the code itself. That’s because /ask uses existing code knowledge as context, for example your code documentation, for answers to queries, resulting in answers that are instant, accurate, and relevant. You probably don’t include the reasons for why you didn’t do something in code, but you can include those nuances in documentation.
/ask Swimm questions like:
- What is X? For example – “what is a Playlist in this repository?”
- Where is X implemented? For example – “where is the editor implemented?”
- How do we X in Y? For example – “how do we send analytics in the editor?”
- Where do we use X? For example – “where do we use the Firestore API in the code?”
- How to add X? For example – “how should I add a new plugin to the system?”
The Swimm Engine
/ask Swimm’s ability to provide accurate, contextual answers beyond what other AI assistants offer is in fact the culmination of research into optimizing developer knowledge sharing in large organizations.
When it comes to engineering knowledge transfer, these challenges permeate all organizations:
- How do we automate the process of capturing knowledge about the codebase?
- How do we ensure it’s up to date and trusted by the team?
- How do we ensure that doc discovery is a natural part of developer workflows?
At Swimm, we have been working on effective knowledge transfer in engineering organizations for years. Turns out, we unknowingly created the optimal context LLMs need to accurately answer questions about your codebase. When you start using /ask, Swimm indexes and runs static analysis on your sources of code knowledge – your codebase as well as existing documentation. That extracted knowledge is then converted into Swimm Generate Documents that describe complex flows in your code in a compact and cohesive way. Auto-sync then keeps those docs up to date. This sequence of events ensures that all of the answers are relevant for your team.
Capture knowledge in the process
Tenured engineers tend to be asked the same questions over and over again. With /ask, developers can automatically capture knowledge and create documentation based on the answers to their questions. That way your team avoids creating documentation that isn’t a value add and over time, documentation coverage increases without sacrificing engineering velocity.
Bringing it all together
Swimm’s mission has always been to make all code easy to understand. We started off as a documentation tool solving the age-old problem of out of date documentation. Building Auto-sync, a CI integration that automatically keeps code snippets in documentation up to date. It turns out that documentation containing code and specific information about that code is the perfect context for an AI coding assistant. The process of asking a question, getting an answer, and then filling in the gaps in the process, creates a continuous feed loop of knowledge about your codebase.
/ask Swimm is currently available on enterprise plans. To learn more, get in touch with the Swimm team.
