kern: Open-Source Tool Maps Codebases Locally for AI Agents via AST Analysis

·3 views

Developer Jayveer Prajapati has released kern, a free open-source tool that builds a local Abstract Syntax Tree index of code repositories to give AI agents like Claude and Cursor precise, context-aware code insights. The tool integrates with AI workflows through the Model Context Protocol and runs entirely on the user's machine, eliminating cloud data exposure and API costs. kern detects over 74 programming frameworks and uses native AST parsing to map call paths, identify dead code, and flag architectural issues without relying on regex guesswork. It also includes a reusable GitHub Action that calculates a mathematical risk score for pull requests and can automatically block high-risk merges before they reach production. Users can set up and index a project with a few commands, and the tool tracks token and cost savings natively through built-in analytics commands.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Tool Sprawl's Real Cost Is Hidden in Complexity, Not Subscription Fees

Organizations typically measure tool sprawl by tallying per-seat SaaS subscription costs, but that figure represents only the most visible fraction of the true expense. Each new tool added to a platform multiplies the number of relationships, integrations, and handoffs that teams must actively maintain with every existing tool. Unlike subscription fees, these hidden costs — such as undocumented tribal knowledge, duplicated functions, and inconsistent policy enforcement — begin accruing the moment a tool enters production and never appear on a renewal invoice. Integration dependencies compound rather than scale linearly, meaning the burden grows far faster than tool count alone would suggest. Finance teams consistently optimize for the wrong question, cutting subscription line items while the deeper operational and architectural costs continue to accumulate unchecked.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AI App Builders Excel at Launch but Fall Short in Post-Deployment Feedback Loops

AI app builders have made it faster than ever to go from idea to deployed product, but critics argue the tools are optimized for the wrong milestone. Real product development begins after launch, when actual user behavior reveals which features matter, where people drop off, and which acquisition channels drive retention. Without post-launch observability, builders lack the context needed to make informed product decisions. The next meaningful leap for AI development tools may not be faster code generation, but rather shortening the cycle between user behavior, actionable insight, and product improvement. Integrating measurement and iteration into the build process could transform AI builders from code generators into end-to-end product development environments.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AI Security Gateway Hits 99.57% Threat Detection After 8.5M-Request Stress Test

A developer published findings from a 36-hour validation sprint after shipping v4.5.0 of an AI security gateway, which initially had five advanced detectors running in alert-only mode rather than actively blocking threats. To determine whether those detectors were safe to enable, the developer built a shadow validation harness using k6 load testing, a mock upstream server, and a Grafana dashboard for real-time monitoring. A progressive stress test ramped up to 10,000 virtual users, generating over 8.5 million requests and peaking at roughly 28,000 requests per second. Across more than 8.1 million benign requests, the false positive rate held at exactly 0%, while the neural-net prompt injection detector achieved a 99.57% true positive rate. With zero false positives confirmed at scale, the developer flipped the L3 neural-net detector from shadow mode to active blocking with a single configuration change.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AWS-Powered Internal AI Assistant Uses RAG Architecture to Unlock Org Knowledge

A developer has built a production-ready internal AI assistant on AWS that allows authenticated staff to query an organization's own documents using retrieval-augmented generation (RAG). The system integrates over a dozen AWS services, including Amazon Bedrock for the AI model, Cognito for authentication, API Gateway, Lambda, and S3 for document storage. Bedrock Guardrails are applied to control both model inputs and outputs, while CloudWatch handles logging and monitoring across the stack. The assistant is deliberately designed around a human review process, generating draft responses for staff rather than autonomously publishing content or taking actions. Terraform is used to provision the entire infrastructure, making the architecture repeatable and suitable for real-world organizational deployment.

0 comments Read more at DEV Community