GitHub - 8Network/8v: 8v is one binary that reads, writes, searches, checks, builds, tests, and formats code, using fewer input and output tokens than the agent's native tools.

You and your AI coding agent — same tool, same rules.

8v is one binary that reads, writes, searches, checks, builds, tests, and formats code, using fewer input and output tokens than the agent's native tools.

Today, 8v supports Claude Code. More agents next.

Install

curl -fsSL https://install.8vast.io | sh

Or build from source:

cargo build -p o8v --release
# binary at target/release/8v

Setup

In any project:

8v init

8v init replaces Claude Code's native Read, Edit, Write, Grep, and Glob with 8v. An MCP server is also available: 8v mcp. Bash stays available for git, processes, and environment.

Commands

8v ls                       # list files and projects in the tree
8v read src/main.rs         # symbol map (functions, types) — fast
8v read src/main.rs:40-80   # exact line range
8v read src/main.rs --full  # whole file
8v search "fn parse"        # regex across the codebase
8v write path:12 "..."      # replace a line
8v write path --insert      # insert, --delete, --append
8v write path --find a --replace b
8v check .                  # lint + type-check + format-check, every stack
8v fmt .                    # auto-format
8v build .                  # invoke the project's build tool
8v test .                   # invoke the project's test runner
8v log                      # what ran in this session
8v stats                    # what failed most often
8v upgrade                  # update 8v itself

Every command accepts --json for structured output.

Stacks

Rust, TypeScript, JavaScript, Python, Go, Deno, .NET, Ruby, Java, Kotlin, Swift, Terraform, Dockerfile, Helm, Kustomize, Erlang. Shell files always go through shellcheck and shfmt.

Benchmark

Each scenario gives Claude a broken codebase and asks it to fix it. Native tools vs. 8v, one variable at a time.

Scenario	Input tokens	Output tokens
fix-failing-test (Rust)	−14%	−42%
fix-go	−21%	−52%
fix-python	−66%	−66%
fix-typescript	−12%	−39%

N=6 per condition. Tests pass 6/6 in every scenario, both arms.

cargo test -p o8v --test demo_agent_benchmark -- --ignored --nocapture --test-threads=1

Workspace

cargo build -p o8v --release   # binary
cargo test --workspace         # full suite
8v check .                     # 8v on itself

Crates: o8v (binary), o8v-core (commands and rendering), o8v-fs (safe filesystem), o8v-process (safe subprocess), o8v-stacks (stack detection and dispatch), o8v-check (lint/type/format orchestration), o8v-testkit (test infrastructure).

License

o8v-fs and o8v-process are MIT. All other crates are BSL-1.1, converting to Apache 2.0 on April 5, 2030.

推荐订阅源

Hacker News - Newest: "AI"