Howdy!

“Where is this function used?, “Which repos still use the deprecated API?, “Can I update this library across 50 microservices without losing my mind?”

I built Code Search to answer these questions and go further - not just finding code, but changing it at scale. It’s a self-hosted platform that indexes your entire codebase and enables bulk operations with automated merge request creation.

What It Does

Search Instantly: Sub-second search across all repositories using Zoekt, the same trigram-based engine Google uses internally. Regex, literal, and structural queries all supported.

Bulk Replace: Find patterns across hundreds of repos, make changes, and automatically create merge/pull requests - all from one command or UI workflow. Preview changes before committing, roll back if needed.

Privacy First: Self-hosted on your infrastructure. Your code never leaves your servers. No telemetry, no external dependencies, complete data sovereignty.

Multi-Host Support: Connect GitHub, GitLab, Bitbucket, and Gitea simultaneously. Mix and match code hosts in a single search.

The Platform

Code Search is a complete ecosystem with multiple entry points:

Web UI (Next.js): Modern interface for searching code, browsing files with syntax highlighting, managing repositories, and executing bulk operations with visual diffs.
REST API (Go): Handles search, repository management, and job coordination. Fully documented with OpenAPI specs.
Indexer Service (Go): Discovers and clones repos from code hosts, then builds searchable trigram indexes. Runs continuously or on-demand.
CLI: Terminal interface for power users and automation. Covered in detail in the Code Search CLI project.

Comprehensive documentation is available at the Code Search Documentation site.

Why I Built This

I started with Hound - a simple, fast code search tool. It worked well, but I kept hitting limitations. No bulk replace operations. No MR/PR creation. No repository management UI. Every time I needed a feature, I had to work around the tool instead of with it.

Sure, there are alternatives: Sourcegraph (requires Kubernetes and enterprise resources), OpenGrok (complex setup, Java-based), grep.app (SaaS, your code leaves your servers). Each has trade-offs I wasn’t willing to make.

I wanted to build something for myself - a tool with exactly the features I needed, on infrastructure I control, with the flexibility to add more capabilities as my needs evolve. So I did.

What I wanted:

Hound’s speed and simplicity, but with bulk operations
Self-hosted on simple Docker containers (no Kubernetes required)
Automated find-and-replace with MR/PR creation
A platform I can extend without fighting the architecture
Full control: my code, my servers, my rules

What I Learned

Building a code search platform from scratch taught me about:

Search Performance: Zoekt’s trigram indexing is brilliant but requires careful tuning. I learned how to optimize index builds, manage memory during indexing, and parallelize operations across multiple repos.

Distributed Systems: Coordinating workers with Redis queues, handling concurrent index updates, and preventing race conditions when multiple indexers run simultaneously.

API Design: Building a REST API that serves both a Next.js frontend and a CLI required careful endpoint design - especially for streaming search results and long-running replace operations.

Database Optimization: PostgreSQL/MySQL sharding strategies for large repo counts, efficient metadata queries, and managing connection pools under heavy load.

Tech Stack

Frontend: Next.js 15, TypeScript, Tailwind CSS, React Query
Backend: Go 1.23+, Chi router, sqlc for type-safe queries
Search Engine: Zoekt (trigram-based indexing), with optional SCIP for symbol-aware search
Storage: PostgreSQL or MySQL (metadata), Redis (job queues, caching)
Deployment: Docker Compose for small setups, Helm charts for Kubernetes
Observability: OpenTelemetry tracing, Prometheus metrics, structured logging

Designed to scale from small teams (single Docker host) to organizations with thousands of repositories (Kubernetes cluster with horizontal scaling).

Status

Actively developed and production-ready - currently indexing over 1,000 repositories across GitHub and GitLab.

Want to know more? Feel free to reach out or check out the Code Search Documentation for setup guides, API references, and architecture details.