Peek is a new Claude Code plugin from early-stage startup gopeek.ai that tries to replace static CLAUDE.md-style markdown instruction files with a dynamic preference-learning system. Rather than requiring developers to manually maintain context files, Peek observes interactions within Claude Code sessions, infers user preferences, and injects relevant memories at appropriate moments. Installation runs through the Claude Code plugin marketplace using the GitHub repository Project-White-Rabbit/peek-claude-plugin; users authenticate via a /peek:login command after setup.
The plugin's Hacker News launch, posted under the title "Show HN: Simple plugin to get Claude Code to listen to you," drew immediate scrutiny over an undisclosed data practice: all user prompts are transmitted to and processed by Peek's servers at gopeek.ai. The detail appeared nowhere in the product's marketing or installation documentation — a community member raised it, not the founders. For developers in enterprise or regulated environments, it's a material concern. Every Claude Code prompt, including proprietary code and confidential context, passes through a third-party server before any memory is formed.
The founder, posting on HN as itsankur, acknowledged the server-side processing and described it as a deliberate choice to enable rapid iteration on data models. The roadmap includes whitelists and blacklists for files and topics, memory export functionality, and possibly self-hosted or local deployments. The founder said API keys, tokens, and PII are actively scrubbed and never stored. Multiple commenters called the lack of upfront disclosure a serious problem regardless, and the thread became as much a debate about developer privacy norms as a product discussion.
Context-persistence tools have become one of the more crowded corners of the LLM developer tooling market. Memory layers, CLAUDE.md management scripts, and session-context injectors have all launched in the past year, and at least one HN commenter flagged Peek as entering an already-saturated space. Gopeek.ai's clearest path to standing out is the self-hosted option the founder floated — though no public timeline for it has been committed.