Karol Kozer
Karol Kozer is an individual solo developer and creator, not a company. He builds agentwhy, a free, local-first tool that reconstructs what AI coding agents and their helpers did with project files, warns when private files were reached, and helps users decide whether to block or track them.
At a Glance
- Individual developers and software teams using AI coding agents
- Teams using Claude Code, Codex or GitHub Copilot
- Companies concerned about agent access to source code, credentials and private files
- Developers building AI agents or tools where users need visibility into file access
AI Tools by Karol Kozer
(1)
agentwhy
AI Agent Secret Access Auditor
Discussions
No discussions yet
Be the first to start a discussion about Karol Kozer
Latest News
agentwhy v0.4.2 released, fixing Step by step read-status and duplicate-file presentation
agentwhy v0.4.1 released with progressive conversation loading and expanded Codex/Copilot settings detail
agentwhy v0.4.0 released with faster page and project/conversation switching
agentwhy v0.3.4 released with expanded blocking behavior and clearer Claude Code stop feedback
Products & Services
An Apache-2.0 open-source npm CLI/package (npx @agentwhy/cli) that runs on the user's computer, reads locally saved Claude Code, Codex and Copilot conversations, and opens a local report showing which private files were reached, by whom, why, and what to do.
The CLI includes report, check, init, protect and notify commands. It can add blocking rules/hooks, track or block private paths, explain helper-agent activity, and identify routes such as searches that bypass simple filename rules.
Kozer's earlier open-source React component for EPG, schedules, live streaming, music events and timelines. It uses a custom virtual view for large datasets and a customizable theme/API; the Planby repository also documents a separate React Native beta package.
Market Position
agentwhy positions itself as a local, retrospective audit and explanation layer for AI coding-agent file access, rather than only a command-level blocker. Its differentiators are reconstructing helper-agent causality from saved transcripts, showing what happened before installation, covering search-based access that simple rules miss, never uploading file contents, and combining reporting with optional block/track controls. It is adjacent to secret scanners, sandboxing and agent-permission tools, but focuses on explaining the complete path from task to file to returned result.
Leadership
Founders
Karol Kozer
Senior Software Engineer / Fullstack and front-end developer; LinkedIn identifies him as a senior software engineer, and his GitHub profile highlights React expertise. Before agentwhy, he created Planby, a React schedule, timeline and EPG component that had more than 1,700 GitHub stars and won first place at WaysAwards 2024.
Executive Team
Karol Kozer
Maker and sole developer of agentwhy
Senior Software Engineer / Fullstack and front-end developer; creator of Planby and an experienced React developer.
Founding Story
Kozer started agentwhy after instructing his AI agent not to open an API-key file, yet a helper agent that the main agent launched searched for the key and returned part of it to the conversation. He found that the relevant evidence was already split across machine-readable conversation and helper records on his computer, so he built agentwhy to join the story together and show what happened and what to change.
Business Model
Revenue Model
agentwhy itself is free and open source. Kozer invites sponsorship to fund maintenance and offers paid company services including hiring, time-based reviews/workshops/talks, ongoing support under a written fixed-monthly-price agreement, and integration into another AI-agent product.
Pricing Tiers
The CLI is free to use and distributed under Apache License 2.0.
The site describes paid consulting, workshops, talks, setup/training and product integration, but does not publish numeric rates.
Target Markets
- Individual developers and software teams using AI coding agents
- Teams using Claude Code, Codex or GitHub Copilot
- Companies concerned about agent access to source code, credentials and private files
- Developers building AI agents or tools where users need visibility into file access
- Auditing what an AI coding agent or helper opened in a software project
- Finding API keys, .env files, SSH material and other private files that reached an AI conversation
- Investigating how a secret moved through a helper-agent task or search result
- Retrospectively reviewing conversations from before agentwhy was installed
- Blocking or tracking sensitive project paths during future Claude Code, Codex or Copilot work
- Team workshops, setup, support and integration consulting for companies using AI coding agents