読書リサーチレーダー投資フレームワーク
ログイン / 新規登録
← 読解に戻る
ニューストピック · GitLab Orbit重大2 分で読めるDisconfirmAI 編集部

GitLab Launches Orbit Context Graph: Internal Tests Show 45x Fewer Agent Retries and 4.5x Lower Token Usage

GitLab has released Orbit, a context graph for the software lifecycle designed to solve AI agents' lack of global visibility. Official tests claim it reduces agent retries by up to 45x and token usage by 4.5x; customer Compare the Market reported an increase in code review comment match rate from 58% (RAG) to 70%.

2026-10-29 08:00 GMT+8

GitLab officially launched GitLab Orbit on October 29, 2026, a "context graph" for the software lifecycle. The product aims to address the inefficiency and hallucinations common in current AI coding agents (such as Claude Code and Codex) when operating within large codebases due to a lack of holistic context. According to GitLab's official blog, Orbit integrates code, merge requests, work items, pipelines, vulnerabilities, and ownership into a single live, queryable view that updates continuously as development progresses.

In GitLab's internal testing, AI agents connected to Orbit required up to 45x fewer retries and consumed 4.5x fewer tokens to complete the same tasks using the same models. These figures suggest that providing structured project context significantly reduces the time and cost associated with agents repeatedly querying backends for missing information.

Beyond internal data, UK price comparison platform Compare the Market provided external validation as a design partner. The company tested four context approaches for its internal AI code reviewer across 79 real merge requests. Results showed that the GitLab Orbit-grounded reviewer matched approximately 70% of expected inline comments, whereas traditional Retrieval-Augmented Generation (RAG) methods achieved only about 58%, performing worse than having no added context at all. Ryan Harvey, Head of AI Engineering at Compare the Market, stated that Orbit enabled the AI code reviewer to truly understand the codebase rather than just the diff in front of it, improving comment placement accuracy and change summaries.

ログイン / 新規登録
読書リサーチレーダー投資フレームワーク

GitLab emphasizes that Orbit is not merely a code indexing tool but a full lifecycle graph covering everything from code changes to deployment, vulnerabilities, and owners. The graph leverages existing GitLab data without requiring connector configuration or additional ETL pipeline maintenance, while preserving original permission structures. Orbit is currently available for Premium and Ultimate users, supporting integration with mainstream AI agents like Claude Code and Cursor via MCP protocol, as well as custom integrations via REST API. Pricing is set at $30 per user per month (in GitLab Credits) for each query pack containing 1,000 queries, with overage billed at approximately $0.03 per query.

ソース

  • about.gitlab.com↗
トピック

GitLab Orbit

GitLab于2026年10月29日正式发布Orbit上下文图谱(Context Graph),旨在解决AI代理在大型代码库中因缺乏全局视野而导致的低效与高成本问题。该工具将代码、合并请求、流水线及漏洞数据整合为单一可查询图谱,使代理能基于…

4 ステップ1 件更新日 2026-11-01 00:33

経緯

  1. 2026-06-10GitLab宣布GitLab Orbit进入公开测试版(Public Beta),同时推出下一代源代码管理等功能,声称代理响应速度提升11倍,令牌使用量减少4.5倍。ir.gitlab.com ↗
  2. 2026-06-10GitLab发布博客文章《Introducing GitLab Orbit》,详细介绍其作为软件生命周期上下文图谱的功能,并引用Compare the Market的测试数据:在79个真实合并请求中,Orbit支持的AI代码审查员评论准确率达70%,优于RAG的58%。about.gitlab.com ↗
  3. 2026-10-29GitLab正式宣布GitLab Orbit一般可用性(GA)。新闻稿指出,自引入以来已有超过3,500家组织使用,运行了超过280,000次查询。IDC研究副总裁Arnal Dayaratna评价称,GitLab凭借全生命周期数据优势,有助于改善结果并管理AI规模成本。morningstar.com ↗

未解決の問題

  • GitLab承诺Self-Managed和GitLab Dedicated版本将在“未来几周”内获得Orbit GA支持,具体发布日期尚未明确。
  • 虽然内部测试显示性能显著提升,但第三方独立基准测试(除Compare the Market外)对Orbit在不同规模代码库中的实际效果验证仍较少。
  • Orbit采用查询包(query packs)定价模式,对于高频使用AI代理的大型企业,其长期成本效益相较于传统按令牌计费模式的对比数据尚不充分。
トピック全文を読む →
リサーチを続ける →