SaaS Metrics
Product Hunt
GitNexus (Akon Labs)
The open-source kernel for coding agents
GitNexus (45k Github stars) is a Knowledge Graph Kernel that unifies every codebase in your org into one source of truth your coding agents can query. It resolves your code into a deterministic graph, so agents get exact callers, imports, and impact instead of embedding guesses, across every repo and SCM you run. On our public benchmark (https://www.akonlabs.com/benchmarks), coding agent runs 51% cheaper with GitNexus connected. It works with any agent over MCP.
View Raw Thread
Developer & User Discourse
[Redacted] • Aug 27, 2026
Congrats on the launch folks!!
[Redacted] • Aug 27, 2026
the "file tree mirrors the disk, not the system" line is a good way to put it - most context tools are really just fancier grep. what I use today is basically Claude Code's own repo-map plus whatever I paste in manually, and it breaks exactly where you'd expect: cross-repo calls, since there's no single view spanning our services. the part I'd want to understand before adopting this though is graph freshness - on a fast-moving repo with agents committing constantly, is the graph updated incrementally per-commit, or is there a reindex lag where an agent could query stale call/import data right after a merge?
[Redacted] • Aug 27, 2026
GitNexus feels like a practical solution to one of the biggest bottlenecks in agentic coding i.e. the context. Huge congrats on the launch, Subham!
[Redacted] • Aug 26, 2026
Hey Product Hunt Fam,
I'm Subham from Akon Labs. We built GitNexus because coding agents spend most of their budget just finding context, not writing code. They read files, grep around, follow imports by hand, and burn tokens rebuilding a mental model of your codebase on every single task. It gets worse the moment that code is spread across multiple GitHub accounts, GitLab, Azure DevOps, and self-hosted enterprise. There's no single view an agent can point at.
An agent is only as good as the model of the world you hand it, and most tooling hands it a proxy. A file tree mirrors the disk, not the system. A per-repo index mirrors the org chart, not the call graph. Embeddings return "who calls this function" as a ranked guess when it has an exact answer.
GitNexus resolves your codebase into one deterministic graph. Nodes are the real entities: symbols, files, functions, services, repos. Edges are the real relationships: calls, imports, implements, deploys. Agents query it and get exact context, not similarity guesses, across every SCM you have.
A few things worth knowing:
- It's open source (45K+ GitHub stars, 1M+ npm downloads), so you can evaluate it in the open
- It makes both closed and open source models cheaper and more capable. Benchmarks : https://www.akonlabs.com/benchmarks
- Works over MCP with Claude Code and your existing agents, no lock-in. Run it managed, or self-hosted and air gapped
I'll be around all day. I'd genuinely love to hear what you're using as your codebase context today and where it breaks first. Ask me anything 🙏
I'm Subham from Akon Labs. We built GitNexus because coding agents spend most of their budget just finding context, not writing code. They read files, grep around, follow imports by hand, and burn tokens rebuilding a mental model of your codebase on every single task. It gets worse the moment that code is spread across multiple GitHub accounts, GitLab, Azure DevOps, and self-hosted enterprise. There's no single view an agent can point at.
An agent is only as good as the model of the world you hand it, and most tooling hands it a proxy. A file tree mirrors the disk, not the system. A per-repo index mirrors the org chart, not the call graph. Embeddings return "who calls this function" as a ranked guess when it has an exact answer.
GitNexus resolves your codebase into one deterministic graph. Nodes are the real entities: symbols, files, functions, services, repos. Edges are the real relationships: calls, imports, implements, deploys. Agents query it and get exact context, not similarity guesses, across every SCM you have.
A few things worth knowing:
- It's open source (45K+ GitHub stars, 1M+ npm downloads), so you can evaluate it in the open
- It makes both closed and open source models cheaper and more capable. Benchmarks : https://www.akonlabs.com/benchmarks
- Works over MCP with Claude Code and your existing agents, no lock-in. Run it managed, or self-hosted and air gapped
I'll be around all day. I'd genuinely love to hear what you're using as your codebase context today and where it breaks first. Ask me anything 🙏