Every instruction is scanned for prompt injection. Every AI-generated edit is checked for vulnerabilities before it touches your code. Every operation leaves an audit trail. Local-first, open source, free.
Edit code without showing you what changed
Most AI agents apply changes directly. You discover what happened after the fact.
No security scanning on generated code
AI can introduce SQL injection, hardcoded secrets, or unsafe patterns — and no one catches it.
Vulnerable to prompt injection attacks
Malicious instructions in comments or files can redirect the agent to do something harmful.
No audit trail for compliance
In regulated industries, you need to prove what AI did and why. Most tools give you nothing.
Public beta — core features work today. Install it, break it, tell us.
Index your codebase into a local SQLite database. Ask natural language questions and get real answers with file paths and code snippets.
Prompt injection scan → AI generates fix → SAST scan → license check → human review gate → audit log. Six stages, every time.
A public, reproducible benchmark: 50 attack patterns across 5 categories. Run in ~50ms from source, no API key. 100% detection in pattern mode. Targeting v0.2.0.
Break down complex changes into step-by-step plans with affected files, dependency analysis, and risk assessment.
Alpha code in development — not yet published to the Marketplace or tested with real users. Targeting v0.2.0. Do not rely on it yet.
All embeddings stay in .codehere-cache/ on your machine. Works fully offline with Ollama. Switch providers with one env var.
Run `codehere index` in your project. It builds a local SQLite database of semantic embeddings. Your code stays on your machine.
Ask questions about your codebase. Plan multi-step changes. Or fix files directly — the safety pipeline runs automatically on every fix.
See exactly what will change. Security scan results, file diffs, risk assessment. Approve to apply, reject to discard.
We built a public benchmark: 50 real-world attack patterns, 5 categories. Run it in 50ms. No API key needed.
Frontier AI labs build powerful models. We build the safety layer between those models and your codebase.
Install. Index. Ask a question. Under 2 minutes. Free and open source. Works offline.