Loading tool details…
Terminal-native AI coding agent that indexes your entire codebase in 2 seconds and makes surgical, multi-file changes from one prompt — an alternative to Claude Code and Cursor Composer.
Terminal-native AI coding agent that indexes your entire codebase in 2 seconds and makes surgical, multi-file changes from one prompt — an alternative to Claude Code and Cursor Composer.
Codebuff (codebuff.com) is an AI coding agent that lives in your terminal, installed with a single npm install -g codebuff. Its differentiator is deep codebase awareness: Codebuff indexes your entire project in about two seconds, then uses that map — including hidden dependencies — to make pinpoint edits across many files while respecting the codebase's existing structure and style. The team claims 100+ seconds faster completions and higher-quality output than Claude Code, backed by their own BuffBench evaluation of 175+ real tasks reconstructed from open-source git commits, scored on completion, efficiency, code quality, and an overall metric across 5-turn conversations. Because it's terminal-based and stack-agnostic, there's no special environment to set up — Codebuff works with whatever tech stack and tooling you already run. Developer testimonials on the homepage emphasize whole-codebase refactors that would take a human 4-8 hours completing in about 30 minutes, and the ability to make sweeping changes in a single prompt rather than the many-step 'composer' flow other tools require. Pricing has two shapes: subscriptions of $100/mo (1× usage), $200/mo (2.5× usage), and $500/mo (7× usage) unlock the 'Strong' mode with all modes and higher usage limits; alternatively, pay-as-you-go bills 1¢ per credit after 500 free credits are granted to GitHub accounts older than 4 months. Credits are consumed proportionally to task complexity, and the UI shows you the consumption before you commit. Codebuff is community-backed and iterated on rapidly by founders who eat their own dogfood daily.
Builds and continuously refreshes a semantic map of the entire codebase, letting the agent answer questions about cross-file behavior (webhook flows, error handling patterns) with grounded citations instead of guessing.
Configurable routing across Claude 4, GPT-5, and Gemini 2.5 so users can pick the right model per task or fall back to a cheaper one when a step doesn't need frontier reasoning.
Runs your existing test suite after edits, reads failures, and iterates until tests pass — with a permission gate on destructive shell commands so it's safe to leave unattended for a diff review.
1¢-per-credit pricing shows an estimated cost before each task runs, giving developers precise control over spend versus flat per-seat SaaS pricing.
Connects to any Model Context Protocol server so the agent can call filesystems, browsers, databases, or internal tools without hand-written function-calling glue.
1¢/credit
$100/month
$200/month
$500/month
Ready to get started with Codebuff?
View Pricing Options →Pricing and features change often. If anything on this Codebuff page is wrong or missing, tell us and include a link to the source. We review every correction.