Sourcegraph

Today, we're announcing Sourcegraph 7.0, a release that marks the beginning of a new chapter for our company and product. Over the past several releases, we've shifted our focus. We're doubling down on being the intelligence layer that developers and AI agents rely on to navigate, understand, and operate on large codebases. sourcegraph.com/blog/a-new-era…

More ways engineering teams manage and transform large codebases: sourcegraph.com/use-cases?&utm…

Code coverage numbers do not mean much if parts of your pipeline are being skipped.

I'm giving a talk at a @mlopscommunity + @qdrant_engine workshop in a couple weeks in Boston on structuring scientific knowledge for agent-driven discovery :) Workshop link: luma.com/19c49kvs

@webuyhomecash graph

Comment "graph" and I'll send you the link. Graphify turns your entire codebase into a visual knowledge graph — code, docs, even video and audio. Free, open-source replacement for Sourcegraph and CodeScene. #github #opensource #developer

@remondimi @ryan0x44 @sqs @AmpCode Excellent points.

@ryan0x44 @sqs @AmpCode Makes sense tbh. @Sourcegraph has been working hard on semantic understanding of code for years and all of that tooling and plugging into the code at a lower level provides a significantly better harness for the agent

I still don't get how @AmpCode smart mode (which uses Opus 4.6 under the hood) can have so much better taste than Claude Code using Opus 4.6, the difference is massive. I mean I understand how, but it's hard to believe it's this big of a difference.

We fine-tuned a small model to convert natural language into Sourcegraph search queries in under 400 ms. Skip the DSL, just describe what you need.

View everything else shipped: sourcegraph.com/changelog/2026…

4/4) Batch Changes for Bitbucket supports OAuth sourcegraph.com/changelog/batc…

What's new in Sourcegraph, and no, we're not fooling around. 1/4) Query assist: search code in natural language

Nice write-up by the @Sourcegraph team on detecting supply chain security for the LiteLLM compromise with their deep search queries: sourcegraph.com/blog/detecting… thanks for the h/t on the Snyk research ✨

Our own @sgjarmak and @trly_dev wrote about the poisoning attack in the LiteLLM Python package. Here's what they found. sourcegraph.com/blog/detecting…

In 2022, we introduced SCIP (pronounced "skip"), a language-agnostic protocol designed for indexing source code. Today, we are excited to announce that it's transitioning to a community-driven open source project! sourcegraph.com/blog/the-futur…

The (most recent) dustup over MCP vs. CLI got me thinking about the gaps I see in the market every day, and particularly the home-grown MCP wranglers/routers/policies that our customers are building in-house. It’s the wild west out there, and there’s zero consensus on the shape of MCP access in the enterprise. sourcegraph.com/blog/mcp-stori…

@ajaynz/the-software-factory-is-here-context-is-still-missing-90a527b0e0fe" target="_blank" rel="nofollow noopener">medium.com/@ajaynz/the-so…

The "Software Factory" is here, but why do AI agents still produce "confident, wrong code"? The companies shipping 1,000 AI PRs a week aren't pulling ahead on model choice. They're pulling ahead due to context. @ajaynz discovers how teams at Stripe and Uber are moving beyond file-level access to build "Context First" architectures. (1/2)

Unpopular opinion: The way we evaluate AI coding agents is broken and agentic benchmarks are sometimes vibes with a spreadsheet. No deterministic test cases. Post-hoc criteria. Arbitrary ground truth. I wrote a practitioner's guide for evaluating @Sourcegraph MCP properly 👇

We're now live in the @GeminiApp CLI extensions marketplace, making integration easier than ever! Link in thread