This video explains how Graft, an open-source terminal tool, reduces token usage and speeds coding agents by building a local knowledge graph of a project’s code. It details installation, CLI/MCP workflows, benchmarks with high-end models, how Graft updates maps incrementally, and limitations (it maps code only, not docs), plus practical examples and measured performance gains.
– Problem: Agents repeatedly resend full conversations and tool results while searching for code, inflating the context window, token costs, and latency.
– Approach: Graft builds a local JSON knowledge graph (nodes and edges) by reading code, recording dependencies, and attaching relevant file locations to prompts or responding on-demand via MCP.
– Benefits: Benchmarks report up to 4× cheaper token usage in best cases; on average tasks ran ~60% faster, with 46% fewer tool calls, 42% fewer tokens, and ~32% lower cost; maps update incrementally as code changes.
– Limitations & workflow: Graft maps code files only (not PRDs or notes), requires project init and a build step, integrates with Claude Code/Codex and MCP workflows, and includes a browser viewer plus optional natural-language pages for clarity.
Quotes:
Graft makes agents four times cheaper in terms of token usage.
Graft builds a knowledge graph mapping project parts and their connections.
A complete revamp of the landing page took less than 2 minutes.
Statistics
| Upload date: | 2026-09-08 |
|---|---|
| Likes: | 3244 |
| Comments: | 143 |
| Fan Rate: | 1.92% |
| Statistics updated: | 2026-10-08 |
