Turbocharge Claude Code, Cursor, Codex, Gemini & every coding agent: faster, cheaper, with contextual understanding specific to your codebase.
TypeScript ai-agentsanthropicclaude-codeclicode-graphcodexcontext-engineeringcursordeveloper-toolsgeminiknowledge-graphllmmcpmcp-serveropen-sourceopenaitree-sitter
docs: hero + benchmark headline — up to 4x cheaper and 4x faster (real-repo Opus results)
Shrish Dwivedi94a5d80039ef
diff --git a/README.md b/README.mdindex 128c6ae2..4958230a 100644--- a/README.md+++ b/README.md@@ -15,16 +15,9 @@<a href="https://scorecard.dev/viewer/?uri=github.com/NanoNets/Graft"><img src="https://img.shields.io/ossf-scorecard/github.com/NanoNets/Graft?style=for-the-badge&label=openssf%20scorecard" /></a></p>-<!-- numbers from a 162-run benchmark (2026-07-22, 2 repos, 3 arms) — see Benchmark below -->+### Up to **4× cheaper** and **4× faster** — with no loss of correctness.-| Metric | With Graft |-|---|---|-| Cost | **32% less** |-| Tool calls | **46% fewer** |-| Latency | **60% lower** |-| Correctness | equal |--**vs. a standard coding session, no Graft at all.**+<sub>Measured on real tasks (real questions + real merged PRs) in popular open-source repos — PocketBase, ollama, Excalidraw. Same agent, same file tools; only the context differs. [Full per-repo numbers ↓](#tested-on-your-popular-repos)</sub></div>@@ -372,9 +365,11 @@ Graft never answered worse than cold, on any corpus. The pull variant gave up mo## Tested on your popular repos-The sweep above measures the mechanism. The real test is whether graft helps an agent **ship real changes**, not just answer questions about code. So we're running it against widely-used open-source repos on the work that actually matters: real merged pull requests, re-implemented from scratch. First up: **[PocketBase](https://github.com/pocketbase/pocketbase)** (Go, ~350 files).+The sweep above measures the mechanism. The real test is whether graft helps an agent **ship real changes** on code people actually run — not just answer questions. So we benchmark it on popular open-source repos: **15 tasks each** — 10 real developer questions + **5 merged pull requests**, each re-implemented from its base commit and scored against the files the maintainers actually changed. Same agent (Claude Opus), same file tools; the only difference is whether graft is wired in.++Across these repos graft runs **up to 4× cheaper and 4× faster**, with no loss of correctness — it reproduces the real merged PRs by touching the same files the maintainers did. Per-repo detail below.-15 tasks, weighted toward real implementation: **5 merged pull requests** re-implemented from their base commit and scored against the files the maintainers actually changed, plus **10 questions** a developer would genuinely ask while working in the repo. Same agent (Claude Opus), same file tools; the only difference is whether graft is wired in.+### PocketBase (Go, ~350 files)| Aggregate over 15 tasks | Standard Claude Code | With graft ||---|---|---|