What is Fuse
Fuse turns a .NET codebase into one token-efficient payload an AI agent reads in a single call, cutting tokens while keeping the public API intact.
You point Fuse at a .NET solution and get back one file: a structured payload that holds the whole codebase, reduced for token efficiency, with a manifest at the top listing every file and its token cost. An agent reads that one payload instead of opening hundreds of files to find its way around.
# install once
dotnet tool install -g Fuse
# fuse a project, keeping the full public API
fuse dotnet --directory ./src --allFused 511 files
Estimated tokens: 366,121 (-21.5%)
cache: 0 hit / 511 miss
Output: AutoMapper_2026-06-20_366k.txtThat run took 511 C# files down by 21.5 percent and kept 99 percent of public types and 100 percent of public methods. The cut is not deletion: the surface an agent reads the code for is still there.
What it does, in one line
Fuse is a .NET-native codebase context optimizer. It collects source files, reduces them, and emits one payload an agent or a developer can consume in a single call instead of reading thousands of files individually.
A single run is a fusion: it takes a source directory and produces one output, either a file on disk or an in-memory payload returned to an AI client.
Why that matters
An AI coding agent spends the start of every task exploring: listing directories, searching, and opening file after file to learn which ones are relevant. On a large solution that discovery burns most of the context window and many slow round-trips before any real work begins. Fuse front-loads that work into one deterministic step and hands the agent scoped, reduced context that fits the window.
Reduction without losing the surface
Across a commit-pinned corpus of real .NET libraries, the default and --all
reductions cut 7 to 40 percent of tokens while keeping 99 to 100 percent of public
types and methods. Skeleton mode, which keeps signatures and drops bodies, cuts 66
to 93 percent for an architecture-level survey.

Every number comes from a harness anyone can rerun against the same pinned commits. See Benchmarks for the method and the full results, including where Fuse loses.
Two ways to run it
Fuse ships from one package in two forms:
- The
fuseglobal tool writes a fusion to a file from your terminal. Start at the Quickstart. - The
fuse serveMCP server exposes eight tools to an AI client such as Claude Code, Cursor, or GitHub Copilot, so the agent requests scoped context directly. See Connect to your AI.
What it is not
Fuse is not a generic file packer that concatenates text, and it is not an embedding index that returns fuzzy chunks. It understands C# structure, preserves the public API, and produces the same output for the same input. The Why Fuse page covers the landscape and the trade-offs in full.
Next
Install the tool on the Install page, then run your first fusion in the Quickstart.
Fuse documentation
Fuse is a .NET-native codebase context optimizer for AI-assisted development. Start here to install it, connect it to your agent, and learn how it cuts tokens while keeping the public API.
Why Fuse, and how it compares
The problem agentic coding tools hit on large codebases, what generic packers and RAG indexers do and where they fall short, and what Fuse does differently.