Graft

Graft is an AI developer tool for building local codebase context graphs for AI coding agents. The source reviewed for this listing is the NanoNets/Graft GitHub repository, plus the public project metadata available during the 2026-08-17 entity creation run. The strongest signal is that the project describes a specific builder workflow rather than a broad assistant promise. The README reports a 50-instance SWE-bench Verified comparison where Graft resolved 33 of 50 instances against 27 for a cold Claude Code baseline, with fewer tool calls, tokens, and wall-clock time. The practical workflow is simple to understand. A developer installs the CLI with npm, runs graft init, builds a local graph for a repository, and lets supported coding agents read that graph during future sessions. A team can start with the documented setup path, run it against a real repository or infrastructure account, and judge whether the output saves repeated manual work. That matters for OpenTools readers because many AI products fail at the handoff between a model response and the actual system a builder needs to operate. Graft is best for software teams using Claude Code, Cursor, Codex, Gemini, or other coding agents on large repositories where repeated codebase exploration wastes time and tokens. It is less useful for nontechnical buyers who want a fully managed support contract on day one. The project gives technical users source code, configuration, and visible behavior they can inspect. That makes the tool easier to test in a small pilot before it touches production data, shared infrastructure, or a team-wide coding workflow. Setup and operations should be treated as engineering work. The README lists npm install -g @nanonets/graft followed by graft init, with npx @nanonets/graft init as an alternative. Teams should read the README, pin a version when possible, and verify what network calls, credentials, logs, and local files are involved. If the tool connects to model providers or cloud accounts, those downstream systems still need normal controls such as scoped keys, approval flows, secret rotation, and cost monitoring. Pricing is based on the public source material, not a sales quote. The repository lists the MIT license. Model-generated summaries may still use provider keys and model spend chosen by the operator. Users may still pay for their own model tokens, cloud resources, or machines. For open-source projects, the free code is only one part of the cost; the larger bill usually comes from APIs, compute, maintenance, and the time needed to keep the workflow safe as the project changes. The main reason to try Graft is that it targets a real bottleneck in agentic software work. Evaluate it with one contained task, record the before-and-after time, inspect the logs, and decide whether the result is easier to review than the old workflow. Because Graft changes how agents see a repository, teams should inspect generated graph files and review diffs before sharing them. If those checks pass, Graft can become a useful layer in a builder stack without forcing a team into a closed platform.