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: SWE-bench numbers over 50 instances
anirudhkumar-nanonetsbb9a9e5d0986
diff --git a/README.md b/README.mdindex 013fadfd..eff821fa 100644--- a/README.md+++ b/README.md@@ -23,9 +23,9 @@| Tool-call savings | Baseline | **+46%** || Token savings | Baseline | **+42%** || Time savings | Baseline | **+60%** |-| Correctness | 69% | **86% (+17 pts)** |+| Correctness | 54% | **66% (+12 pts)** |-<sub>Efficiency is a 162-run controlled benchmark (same agent, same file tools, only the context differs). Correctness is **SWE-bench Verified**, graded by the official harness — graft resolved 86% of instances tested against Cold Claude Code's 69%. The "up to 4× cheaper / 3× faster" figures are the biggest single-task wins from a separate real-repo sweep (PocketBase, ollama, Excalidraw). [Efficiency method ↓](#benchmark) · [SWE-bench ↓](#swe-bench-verified) · [Per-repo numbers ↓](#tested-on-your-popular-repos)</sub>+<sub>Efficiency is a 162-run controlled benchmark (same agent, same file tools, only the context differs). Correctness is **SWE-bench Verified**, graded by the official harness — graft resolved 66% of instances tested against Cold Claude Code's 54%. The "up to 4× cheaper / 3× faster" figures are the biggest single-task wins from a separate real-repo sweep (PocketBase, ollama, Excalidraw). [Efficiency method ↓](#benchmark) · [SWE-bench ↓](#swe-bench-verified) · [Per-repo numbers ↓](#tested-on-your-popular-repos)</sub></div>@@ -106,7 +106,7 @@ Graft builds that understanding **once** and writes it into your repo as a folde<p align="center"><picture><source media="(prefers-color-scheme: dark)" srcset="assets/graft-cold-vs-graft-dark.png">- <img src="assets/graft-cold-vs-graft.png" alt="The same task, 'fix the auth bug', run two ways. A cold Claude Code session re-reads the repo and wanders file to file; Claude Code + graft loads its map once and rides the hooks to one clean pass. With Graft: 46% fewer tool calls, 42% fewer tokens, 60% less time, +24% more SWE-bench instances resolved." width="880"/>+ <img src="assets/graft-cold-vs-graft.png" alt="The same task, 'fix the auth bug', run two ways. A cold Claude Code session re-reads the repo and wanders file to file; Claude Code + graft loads its map once and rides the hooks to one clean pass. With Graft: 46% fewer tool calls, 42% fewer tokens, 60% less time, +22% more SWE-bench instances resolved." width="880"/></picture></p>@@ -136,17 +136,16 @@ Graft never answered worse than cold, on any corpus. The pull variant gave up moThe sweep above is our harness measuring our mechanism. So we ran the industry-standard one too — **SWE-bench Verified**, real GitHub issues from real repos, graded by the official `swebench` harness. No judge model, no similarity score: your patch is applied, the maintainers' own tests are run, and you either flip the failing test without breaking the passing ones or you don't.-**36 instances**, same model on both arms — **Claude Sonnet 5** — same Docker images, same turn limits. The only difference is whether graft is wired in.+**50 instances**, same model on both arms — **Claude Sonnet 5** — same Docker images, same turn limits. The only difference is whether graft is wired in.| Correctness & efficiency | Cold Claude Code | Claude Code with graft | Improvement ||---|---|---|---|-| Correctness | 25 / 36 (69%) | **31 / 36 (86%)** | **+17 pts** |-| Tests passed | 6,523 | **6,543** | **+20** |-| Token savings | 107.0M | **85.0M** | **+21%** |-| Cost savings | $39.34 | **$31.93** | **+19%** |-| Tool-call savings | 998 | **759** | **+24%** |-| API-request savings | 1,777 | **1,377** | **+23%** |-| Wall-clock savings | 8,721s | **6,488s** | **+26%** |+| Correctness | 27 / 50 (54%) | **33 / 50 (66%)** | **+12 pts** |+| Token savings | 142.0M | **109.4M** | **+23%** |+| Cost savings | $52.34 | **$42.43** | **+19%** |+| Tool-call savings | 1,370 | **1,031** | **+25%** |+| API-request savings | 2,455 | **1,875** | **+24%** |+| Wall-clock savings | 13,094s | **8,922s** | **+32%** |### Where graft fixed what cold missed@@ -190,11 +189,11 @@ The sweep above is our harness measuring our mechanism. So we ran the industry-s| `sphinx-8120` | Passed: 45 / 45 tests | **Passed: 45 / 45 tests** | **93%** | **75%** || `django-15103` | Passed: 19 / 19 tests | **Passed: 19 / 19 tests** | **96%** | **80%** |-graft resolved **31 of 36 instances** against Cold Claude Code's 25 — and got there with 24% fewer tool calls, 21% fewer tokens, and 26% less wall-clock time. Every correctness win has the same shape: the baseline patches one file and misses its siblings. On `django-11532` it patched 1 of the 5 files the fix requires and broke 18 previously-passing tests, twice over. On `django-16263` it patched 1 of 4 and scored 102 / 103. graft found the rest — and on `django-16263` did it in half the tokens and half the time.+graft resolved **33 of 50 instances** against Cold Claude Code's 27 — and got there with 25% fewer tool calls, 23% fewer tokens, and 32% less wall-clock time. Every correctness win has the same shape: the baseline patches one file and misses its siblings. On `django-11532` it patched 1 of the 5 files the fix requires and broke 18 previously-passing tests, twice over. On `django-16263` it patched 1 of 4 and scored 102 / 103. graft found the rest — and on `django-16263` did it in half the tokens and half the time.Two harnesses, two claims: the controlled sweep says graft is cheaper and faster, SWE-bench says it's also more correct.-<sub>Correctness and tests over all instances; tokens, cost and calls over the instances both arms resolved, for a like-for-like comparison. Official SWE-bench Verified images and official `swebench` 4.1.0 grader, native x86_64.</sub>+<sub>Correctness over all instances; tokens, cost and calls over the instances both arms resolved, for a like-for-like comparison. Official SWE-bench Verified images and official `swebench` 4.1.0 grader, native x86_64.</sub>---diff --git a/assets/graft-cold-vs-graft-dark.png b/assets/graft-cold-vs-graft-dark.pngindex f38d4dd1..07433c92 100644Binary files a/assets/graft-cold-vs-graft-dark.png and b/assets/graft-cold-vs-graft-dark.png differdiff --git a/assets/graft-cold-vs-graft.png b/assets/graft-cold-vs-graft.pngindex c5ff19b9..c2b02a0e 100644Binary files a/assets/graft-cold-vs-graft.png and b/assets/graft-cold-vs-graft.png differ