Every engineer eventually hits the same wall: the answer to "where do we already do this" lives in a repo that isn't open in their editor. Your IDE assistant only sees the current checkout, which is the wrong scope the moment a bug touches three services and a Terraform module. I went looking for what actually solves that in 2026, and the market looks nothing like the 2024 comparison posts still circulating.
I wrote up the full research on DevToolLab, but the short version: bloop, the Rust-based AI code search engine with 9,500 GitHub stars, is gone. Archived in January 2025, company shut down in April 2026, domain now parked. Meanwhile the free tier of this category got surprisingly good. I indexed a shallow clone of kubernetes/kubernetes, 31,300 files, with Zoekt on a laptop. 6.03 seconds to index, 193ms for a filtered regex query across the whole thing.
The market split in two directions
Sourcegraph went upmarket. Its pricing page now shows one plan, Enterprise, starting at $16,000/year, "contact sales" instead of a signup button. The $9-19/month individual Cody plans are gone entirely.
Greptile went sideways. It built its name on natural-language Q&A over a codebase, but its homepage now says "The AI Code Reviewer," and the docs are all pull request workflows. The underlying code graph is still there, it's just feeding review agents instead of a search box.
The thread connecting both moves: agents, not people, are now the main consumer of code search. Sourcegraph, Sourcebot, and Greptile all ship MCP servers, because the thing asking "where is this pattern used" is increasingly Claude Code or Cursor.
Sourcegraph Deep Search
Sourcegraph's 2026 pitch is Deep Search, an agent that investigates across every repo in your org, follows code and its commit history, and cites the files and diffs behind each answer. It's still the only thing here that can aggregate across a whole portfolio, "every repo still on a vulnerable log4j" as one cited answer instead of a results page. The catch is there's no self-serve path anymore. If you just want free public code search, sourcegraph.com/search still does regex and diff search across open source repos.
Greptile, now a reviewer not a search box
Greptile builds a graph index of files, functions and dependencies and runs agent swarms over each pull request, and that index is the same asset that used to power its Q&A product. It's a genuinely strong review tool now, used by 22,000+ teams, with an MCP server into Claude Code, Cursor, VS Code and Codex. Just don't buy it expecting a search UI. The free tier covers one developer with 50 credits/month; paid seats are $30/month.
The free and self-hosted options got good
This is where most teams should actually start. Sourcebot is the direct Sourcegraph-price answer, npx setup-sourcebot or one Docker image, indexing GitHub/GitLab repos with regex, symbol search and an "Ask Sourcebot" layer with inline citations. It's built on its own fork of Zoekt, which is why the underlying search feels familiar. Free for the basics, $20/user/month for the AI answer layer and SSO.
Zoekt itself, the trigram engine Sourcegraph maintains and open-sources, is the one worth trying first regardless of budget. Apache 2.0, no web UI at all, that's the point, it's meant to be embedded:
go install github.com/sourcegraph/zoekt/cmd/zoekt-git-index@latest
go install github.com/sourcegraph/zoekt/cmd/zoekt@latest
git clone --depth 1 https://github.com/kubernetes/kubernetes.git kube
zoekt-git-index -index ./index -branches HEAD kube
That indexed a 396MB clone (31,300 files) in 6.03 seconds. A follow-up query:
zoekt -index_dir ./index -l 'file:\.go$ hmac\.(Equal|New)'
7 files back in 193ms cold, 46ms warm. One gotcha worth saving you the debugging time: write the regex bare, not slash-delimited like sourcegraph.com expects, or the CLI silently returns nothing.
For a Java-heavy on-prem shop, OpenGrok is still shipping regular releases (1.14.17 in August 2026) after two decades, no AI layer, but real cross-reference precision.
How to actually decide
Skip the demos and run this in ten minutes: index your biggest repo with Zoekt, then ask it the three questions your team argued about last week. If regex plus a file filter answers them, you don't need a paid natural-language layer at all, you need an index, and that's free. If the questions need behavior traced across services, that's what Deep Search or Ask Sourcebot are actually for. Either way, wire the MCP server into your coding agent and re-run the same questions from inside it, since that's increasingly where the queries will actually happen. I go through the full decision framework with pricing tiers by team size on DevToolLab.
The bloop lesson generalizes past this one post: before adopting anything whose value is an index it maintains for you, check the last commit date. That single check would have flagged bloop a year before its domain lapsed.
References
- Best AI Code Search Tools in 2026: Sourcegraph vs Open Source - the original article, with the full comparison table and a step-by-step evaluation guide
- GitHub Repo Stats - check stars, forks and issue activity before betting on an open-source search engine
- Regex Explainer - break down a dense search pattern before running it across a whole repo
- Zoekt on GitHub
- bloop's "Goodbye" post
Top comments (0)