Sage AI - Your self-generating, self-maintaining code knowledge base

Hi everyone, we’re Akhil and Joshua and we're on a mission to accelerate expertise and comprehension of any codebase by providing a self-generating, self-maintaining code knowledge base.

🧨 The Problem

The story of every software team is a constant struggle with knowledge friction and maintaining expertise in the code and surrounding systems.

These are the perennial issues we’re solving:

• Staleness: Most documentation is perpetually stale and it slows development to keep it fresh.
• Complexity: Code becomes more difficult to maintain and improve as it increases in age and size. This is the main reason enterprises with extremely large, old, and complex code can’t move fast.
• Fragmentation: Code information is unversioned and unlinked across disparate platforms, so it’s often unclear what your Confluence article is referring to and whether it’s still applicable.
• Discoordination: Information silos impede collaboration and lead to unnecessary repetition of work. Cross-team features and bugs are exponentially harder to implement and resolve.
• Underutilization: Engineers are expected to take 8-12 months to fully onboard and usually experience a “sink-or-swim” strategy for ramping up a new team.
• Overdependence: Core knowledge is continually lost as employees turnover. Raise your hand if you’ve ever needed to touch code that’s written by an engineer who left in 2014. 🙋

Whether you’re a newly hired engineer, just moved teams, or need to touch code you’ve never seen before, it’s the steepest learning curve in software engineering. This is the skeleton-in-the-closet of software: your average engineer spends more than 40% of their 2-year-tenure “onboarding.”

🎉 The Solution

Sage is changing this reality. We take your code and automatically generate a symbol-level knowledge base that self-updates to remain forever fresh.

First, we integrate with source control to pull your code and generate a symbolic graph representation of functions, classes, interfaces, types, etc., automatically determining the relevant context for each symbol.

We then use LLMs to annotate and propagate documentation via traversals of the symbol graph, saturating knowledge across multiple traversals.

We're also exposing a configurable, deeply contextualized chat system that can derive higher-level purposes and objectives of program symbols and modules from its own knowledge base.

In the future, we'll let your engineers themselves improve the system, and integrate and link current knowledge systems such as Slack and Confluence. As its comprehension of your codebase grows, the quality of every downstream task improves automatic coding, testing, refactoring, bug fixing, vulnerability detection, and so on.

Our ask

• If you are a software engineer and have struggled with documentation or onboarding in the past, sign up to try the platform at https://sage-ai.dev. We would love to get your feedback!
• If you know any engineers or engineering managers who might be interested in using our platform with their teams, let us know at founders@sage-ai.dev
• If you are developing AI systems that deal with codebases, let us know if you are interested in consuming docs and symbol graphs as an API.
• Share any additional feedback in the comments!

Sage AI Team (Akhil and Joshua)