Cursor Launches Origin, a Git Hosting Platform Built for AI-Driven Development

·6 views

Cursor has announced Origin, its own git hosting and code collaboration platform designed for teams and AI agents. The product covers repository hosting, code review, and collaboration workflows, targeting a gap in the AI coding pipeline that Cursor previously left to services like GitHub and GitLab. Origin is reported to be connected to Cursor's earlier acquisition of Graphite, which specialised in high-volume pull request workflows. The platform is described as built for the speed at which AI agents generate and push code, which Cursor argues exceeds what traditional forges were designed to handle. A public waitlist is now open at cursor.com/origin, with a general release expected this fall; pricing and full documentation have not yet been published.

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)

Developer finds 8 incorrect WCAG criteria in own accessibility scanner after audit

A developer building a WordPress accessibility scanner discovered that eight of its 25 automated checks referenced incorrect, outdated, or unrelated WCAG criteria, after a reviewer initially flagged three errors. Among the issues, the scanner cited WCAG 4.1.1, a criterion that was fully removed in WCAG 2.2, and misclassified several best practices as binding success criteria. Following the audit, the developer corrected all eight rules, distinguishing genuine WCAG 2.2 success criteria from best practices in both the codebase and product documentation. The scanner's marketing copy was also updated to accurately reflect 18 WCAG-mapped checks and 7 best-practice checks, rather than the previous blanket claim of 25 WCAG controls. The developer published the findings as a self-disclosed account, noting that overstating automated test results can mislead accessibility declarations and reports.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Voiden Stores API Specs, Tests, and Docs in a Single Markdown File

Voiden is an offline-first, Git-native API workspace that consolidates API specifications, tests, and documentation into a single plain Markdown file stored directly in a project's repository. Unlike cloud-based tools such as Postman or Insomnia, it requires no account creation and collects no telemetry, addressing the common problem of fragmented API tooling across teams. The tool supports REST, GraphQL, gRPC, and WebSockets, and allows API requests to be written as reusable composable blocks rather than duplicated entries. Because everything lives in a plain file, documentation updates and code changes can be committed and reviewed together in the same pull request, reducing the risk of stale docs. Voiden also supports direct import from Postman, Insomnia, and OpenAPI specs, lowering the barrier to migration for existing projects.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Chrome DevTools MCP 1.7 adds AI-friendly tools to diagnose browser memory leaks

Chrome DevTools MCP version 1.7, released on August 10, introduces new memory-debugging capabilities for AI coding agents working in tools like Codex, Claude Code, Cursor, and Copilot. The update adds a get_heapsnapshot_object_details tool that lets agents inspect specific heap nodes without parsing raw multi-gigabyte snapshot files. Snapshot summaries now include native context data, and the built-in memory-leak skill has been updated to route analysis through native MCP tools including comparisons, retaining paths, and dominator views. The release also fixes two cleanup bugs where heap-snapshot workers were not properly disposed during context teardown or failed snapshot loads, reducing the risk of confusing debugger overhead with actual application memory growth. Developers are advised to follow a three-snapshot protocol — baseline, target, and final — repeating the suspected action ten times before drawing conclusions.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Selecting the Cheapest AI Model Does Not Guarantee the Cheapest Execution Cost

Developers building AI agents often assume routing tasks to lower-cost models will keep expenses predictable, but actual execution costs frequently diverge from those estimates. A single agent run can span multiple model calls across providers like Anthropic, OpenAI, and Groq, each with different per-token pricing for input and output. Fallbacks triggered by rate limits or latency spikes can silently multiply costs — for example, switching from Haiku to Sonnet for the same task raises input costs nearly fourfold. Unpredictable output lengths compound the problem further, since a verbose model response can cost ten times more than a brief one without any visible warning. Experts recommend logging the actual model selected and real token counts after each call, then computing costs at execution time using published per-million-token rates to replace guesswork with measurable data.

0 comments Read more at DEV Community