Multi-File Editing
Describe a single task and ANVIL coordinates edits across every affected function. Changes are ordered so dependent functions update correctly, and each edit is verified independently before it is applied to your codebase.
ANVIL
AI-Assisted Coding for Air-Gapped and Data Sensitive Networks
Intelligent code editing and generation that runs entirely on local hardware. No cloud. No data exfiltration. No compromise on capability.
Local-Only
Runs on a single workstation GPU. No network dependencies.
Model-Agnostic
Works with any local model, from 7B to 70B+. Swap with a config change.
Structurally Intelligent
Understands your codebase's architecture, not just its text.
See It Work
Watch ANVIL take a natural-language task, identify the affected code, and execute validated edits across multiple files — entirely offline.
anvil-cli
$
This demonstrates the CLI interface. The VS Code extension provides the same capability with inline diffs and accept/reject UI.
85.9%
Sub-edit success rate on a 7B parameter model running on consumer GPU hardware.
Standard local model methods using the same 7B model achieve approximately 11% on the same tasks. ANVIL's orchestration closes the gap between local inference and cloud-hosted AI tools — without sending a single byte off the machine.
Capabilities
ANVIL handles multi-step coding tasks that single-prompt AI tools cannot — coordinating edits, validating results, and retrying failures automatically.
Code Generation
New modules and services scaffolded from the patterns already present in your codebase. ANVIL learns the structure of your existing code — naming conventions, file layout, function signatures — and replicates it for new components, then fills in the implementation automatically.
Intelligent Task Understanding
Describe tasks in natural language without memorizing function names or file paths. ANVIL identifies what you are referring to from your project's vocabulary and determines the full scope of the change — which files, which functions, which callers need updating.
Structural Validation
Every edit is verified before application: syntax correctness, scope enforcement confirming the model only changed what it was asked to change, and signature consistency across dependent functions. Failed edits are retried with specific error feedback — not vague "try again" prompts.
Model Flexibility
Configurable model profiles for different deployment scenarios. A 7B model on a developer's workstation for daily use. A larger model on a shared appliance for complex tasks. Switch between profiles without pipeline changes — same orchestration, different inference backend.
Deployment Flexibility
Single-workstation mode for individual developers: everything on one machine with a GPU. Appliance mode for teams: inference and orchestration on a server, developers connect over the local network. Both modes are fully air-gapped with zero external dependencies.
Who This Is For
Defense & Intelligence
Classified software development, SCIF environments, air-gapped networks. Programs where cloud-based AI tools are prohibited and developers currently have no AI assistance at all. ANVIL brings modern AI-assisted coding to the environments that need it most.
Regulated Industry
HIPAA-covered development, financial services, legal tech — any organization whose compliance posture prohibits sending source code to third-party APIs. Get AI coding assistance without changing your security architecture or risking data exfiltration.
Supported Languages
TypeScript / JavaScript|Roadmap: Python, C/C++, Rust
Secure AI Coding
Interested in deploying ANVIL in your environment?
Whether you operate in a classified enclave or a compliance-restricted enterprise, we can walk you through deployment options and a live demonstration.