GPS.

GPS — Repo Memory for Coding Agents

GPS is a persistent memory layer that teaches coding agents how your repository works once, then surfaces relevant context automatically when code is touched. It eliminates repetitive explanations by storing conventions, best practices, decisions, and lessons learned—making every future edit smarter.

Product Highlights

• Symbol-Anchored Memory: Lessons attach to specific code symbols and surface only when relevant, keeping context precise and noise-free
• Multi-Agent Support: Works identically across Claude Code, Codex CLI, Cursor, and shell-based agents with native integrations
• Automatic Capture: Lifecycle hooks capture corrections and preferences from session transcripts without manual logging
• Decision-Ready Briefs: gps prepare delivers structure, tests, invariants, notes, and risk assessment in one command
• Closed Feedback Loop: Session distillation automatically promotes recurring lessons into durable invariants

Use Cases

• Onboarding New Developers: GPS preserves tribal knowledge—architectural decisions, edge cases, and gotchas—so new team members don't repeat past mistakes
• AI-Assisted Development: Agents receive contextual briefs before editing, reducing hallucinations and repo-specific errors
• Knowledge Preservation: Hard-won lessons from incidents and PRs become searchable, queryable assets rather than lost Slack threads
• Code Review Preparation: Generate structured impact analysis and test recommendations before submitting changes

Target Audience

GPS is built for engineering teams using AI coding assistants who need their agents to understand repository-specific context—particularly valuable for complex codebases with accumulated technical knowledge and teams scaling AI-assisted development practices.