Make every code change predictable and verified.
The local-first AI engineering workbench that maps codebase architecture across major languages (TS/JS, Python, Go, Rust, Java, C/C++), prunes junk and auto-fixes code with instant rollback, audits change risk, and self-heals via test verification — with zero telemetry leakage and full developer control.
Bring ANVIL into your development environment
Run instantly with npx or install globally via your preferred package manager. Requires Node.js 22+.
npm install -g anvil-engineering-workbench
.anvil/ folder.Engineering evidence, not mystery automation.
Explore the four core facets of ANVIL. Each engine gathers concrete evidence to guide your decisions rather than modifying your files silently.
Built for every step of the software lifecycle.
From initial codebase discovery to post-merge verification, ANVIL provides purpose-built tools that give you structural clarity.
Project Discovery
Scan languages, manifests, frameworks, scripts, and dependency counts before touching a single line of code.
anvil inspectArchitecture Mapping
Map internal module relationships, external package imports, entry points, and circular dependency rings.
anvil map · deepAST Symbol Analysis
Parse TypeScript syntax trees to extract functions, classes, interfaces, and transitive blast-radius paths.
TypeScript ASTChange-Risk Assessment
Correlate working-tree git changes with downstream architectural hotspots and recommended test suites.
anvil gitReview History
Save code review snapshots, inspect individual findings, and compare reviews across commits over time.
anvil review-historyDry-Run Patch Validation
Validate proposed unified diffs for target file paths, hunk syntax, and source context without applying them.
anvil patch-checkVerification Engine
Run named project test or build scripts deliberately, capturing standard outputs, exit codes, and timestamps.
anvil verifyRegression Intelligence
Compare saved verification outcomes to identify regressions, recoveries, persistent failures, and stable runs.
anvil regressionEngineering Health Score
Synthesize quality, verification, and dependency evidence into an honest, weighted engineering health metric.
anvil healthDependency Intelligence
Inspect npm package manifests, lockfiles, and audit reports for vulnerability flags, outdated ranges, and licenses.
anvil depsRemediation Planner
Prioritize critical dependency findings and generate reviewable upgrade strategies without silent package installs.
anvil remediateLocal Web Dashboard
Launch an interactive local control center with live SSE broadcasts, interactive graphs, and run inspectors.
anvil dashboardAutomated Junk Cleaner & Fix
Scrub OS residue, partial downloads, error logs, and multi-language caches, with safe unused-import auto-repair.
anvil clean · fixSelf-Healing Doctor Wizard
Automate environment health diagnostics with one-command scaffolding of .anvil directories, .gitignore, and .env.
anvil doctor --fixHow ANVIL works under the hood.
A decoupled, local-first pipeline designed to process real project trees without sending your proprietary source code to third-party endpoints.
Integrates directly with your AI engineering stack.
Connect ANVIL's local read-only MCP server to your AI editor or invoke specialized CLI workflows from automated pipelines.
Claude Code
Register ANVIL as a local stdio MCP tool server so Claude can map architecture, search symbols, and audit codebase risk.
Cursor IDE
Add ANVIL to your project's .cursor/mcp.json for instant repository context inside Composer
and Chat.
Ollama
Point ANVIL to your local Ollama endpoint for zero-cloud, 100% private review suggestions and diff generation.
OpenAI · Anthropic · Gemini
Configure your own API keys for optional high-reasoning patch proposals and security analysis.
A repeatable, disciplined loop
From unfamiliar legacy repositories to safely verified pull requests.
Understand
Inspect frameworks, package manifests, scripts, and build setup before modifying any source code.
Map
Trace module dependencies, AST symbols, entry points, and structural hotspots across files.
Review
Analyze Git uncommitted diffs, assess downstream blast radius, and validate proposed unified patches.
Verify
Execute named test scripts, save deterministic outcome logs, and catch regressions before merging.
Everything you need to know.
Does ANVIL automatically apply changes to my code?
No. By design, ANVIL enforces human approval for consequential actions. Patch validation is a dry run that inspects diff structure, and remediation generation produces a plan rather than silently modifying dependencies.
Is ANVIL's local dashboard a hosted web app?
No. The dashboard runs 100% locally on your machine via Node.js and binds to localhost. Your repository source code, AST index, and git history never get uploaded to any external server.
Which languages are supported for deep AST analysis?
TypeScript and JavaScript have native, deep AST and symbol-level analysis. For other languages (Python, Go, Rust, Java), ANVIL provides manifest parsing, configuration audits, script runners, and git change-risk heuristics.
How do I connect ANVIL with Claude Code or Cursor?
ANVIL provides a built-in MCP server (`anvil mcp` or through the CLI). Simply add it to your Claude Code or Cursor configuration file to grant AI assistants read-only access to ANVIL's repository intelligence tools.
What is required to run ANVIL?
ANVIL requires Node.js version 22 or higher. It works seamlessly across macOS, Linux, and Windows PowerShell.
Is ANVIL completely free and open source?
Yes. ANVIL is licensed under the permissive MIT License. You have complete freedom to inspect, fork, contribute, and use it in commercial or open-source projects.
Ready to bring clarity to your codebase?
Install ANVIL in seconds or explore the complete open-source repository on GitHub.