HeimWall.ai
HeimWall is a macOS application that detects secrets, PII, and proprietary code as engineers type into AI coding assistants like Cursor, Claude Code, and Copilot. It masks sensitive content on-device before it is recorded, writes tamper-evident audit logs, and generates a weekly AI safety score per engineer and team. The tool monitors multiple AI tools from one installation, including personal-account sessions.
- Cybersecurity
- Developer Tools
- Artificial Intelligence
- Software Only
- Data & Analytics
Funding
Founders
Founder details are not available yet.
Product
Problem
Engineers routinely paste API keys, customer records, and proprietary code into AI coding assistants like Cursor, Claude Code, and Copilot, often without realizing it. Traditional data-loss-prevention tools were built for email and file uploads, not agentic chat, and their block-first approach slows developers down so much that teams disable them. Each AI tool's native controls only monitor that single tool, leaving organizations blind to cross-tool leak signals.
Solution
HeimWall is a macOS app that runs locally on an engineer's laptop and intercepts text being typed into Cursor, Claude Code, Copilot, and ChatGPT Desktop. It detects secrets, PII, and proprietary code patterns across 47 rules, masks the sensitive content before anything is recorded, and writes a tamper-evident audit log for team review. The prompt still travels to the AI tool unchanged; HeimWall detects and warns but never sits in the send path. The agent also produces a weekly AI safety score per engineer and per team, which combines severity-weighted event volume, tool-surface coverage, temporal trends, and per-engineer-week normalization into a single 0–100 number that managers can monitor over time.
Target Audience
Primary customers are engineering leaders and security teams at organizations where developers use AI coding assistants, who need visibility into cross-tool leak surfaces without disrupting developer velocity.
Features
- On-device detection engine with 47 rules covering secrets, PII, and proprietary code patterns
- Standalone macOS agent that monitors multiple AI tools (Cursor, Claude Code, Copilot, ChatGPT Desktop) from one installation, including personal-account sessions
- Tamper-evident audit logging with a security clause prohibiting use of score data in performance reviews, promotions, or compensation decisions
- Proprietary AI safety score (0–100 weekly composite) built from severity-weighted event volume, tool-surface coverage penalties, exponential decay on event age, and active-engineer-week normalization
- Detection coverage panel rendered alongside the score to expose false negatives when rules go silent or engineers shift to uninstrumented tools
- Benchmarked alert-firing rate of 1.12% on 27,075 real developer prompts from the DevGPT dataset, with manual triage confirming three live-format API keys and 49 personal email addresses