Skip to content
GitHub·

🤖Graft Cuts Costs and Speeds Up Coding Agents by 4x

Coding agents just got cheaper and faster with Graft

TL;DR

Graft, a new tool that builds contextual understanding of codebases, is up to 4 times cheaper and 3 times faster than traditional coding agents. It reduces tool-call savings by 46%, token savings by 42%, and latency savings by 60%. Graft's graph-based approach ensures deterministic results without model calls.

Graft has emerged as a game-changer in the world of coding agents, boasting up to 4 times cheaper costs and 3 times faster performance compared to traditional tools. This is huge for developers looking to optimize their workflows and reduce expenses. Graft achieves this by building a deterministic graph from markdown files that explain each part of a system and how it connects to others. In benchmarks with 162 runs across two repos, Graft showed significant improvements: reducing tool-call savings by 46%, token savings by 42%, and latency savings by 60%. The graph is built in two passes, both powered by language models, ensuring correctness without model calls.

Graft Cuts Costs and Speeds Up Coding Agents by 4x — GitHub

Key Points

1

Graft reduces tool-call savings by 46%, token savings by 42%, and latency savings by 60%

2

In benchmarks, Graft resolved 33 of 50 instances against Cold Claude Code's 27

3

The graph is built in two passes: summarizing source files into descriptions (first pass) and grouping summaries with typed links (second pass)

4

Graft uses 25% fewer tool calls, 23% fewer tokens, and takes 32% less wall-clock time than Cold Claude Code

5

Each node explains what a part of the system does and how it connects to others

Why It Matters

If you're using coding agents in your development workflow, Graft's cost savings and performance boosts could be a game changer. For instance, teams working on large codebases with complex interdependencies will see significant improvements in efficiency and budget management. The deterministic nature of Graft's graph ensures consistent results without the need for constant model calls.

GraftCoding AgentsCost ReductionPerformance Boost

Frequently Asked Questions

Why does this matter?

If you're using coding agents in your development workflow, Graft's cost savings and performance boosts could be a game changer. For instance, teams working on large codebases with complex interdependencies will see significant improvements in efficiency and budget management. The deterministic nature of Graft's graph ensures consistent results without the need for constant model calls.

What happened?

Graft, a new tool that builds contextual understanding of codebases, is up to 4 times cheaper and 3 times faster than traditional coding agents. It reduces tool-call savings by 46%, token savings by 42%, and latency savings by 60%. Graft's graph-based approach ensures deterministic results without model calls.

Comments

Subscribe to join the conversation...

Be the first to comment

Enjoyed this article?

Get it daily. 7am. Free. Reads in 5 minutes.

Join 3,005 builders reading daily.

Also get