Tony built Grov to solve a core problem with current AI coding assistants: they operate in isolation. When a developer closes a session, the agent's high-level reasoning and architectural decisions vanish. A teammate starting fresh must re-derive everything from scratch, wasting time and tokens on redundant exploration.

Grov acts as an open-source context layer, giving team agents shared, persistent memory. It captures decisions at the grain of reasoning—like choosing JWT for auth because it's stateless for scaling—rather than just storing documents. Old decisions are marked as superseded, creating an audit trail of architectural evolution.

The system uses Git-like branches for memory, isolating experimental approaches until they're ready to merge. A two-stage injection strategy optimizes context windows: a semantic/keyword search returns lightweight summaries first, expanding only when requested. This can reduce token usage by 50-70% per session compared to raw context dumping.

For teams, the result is that one agent's learning becomes instantly available to others. Instead of re-investigating why Postgres was chosen over Redis, the agent simply knows. Grov is free for individuals and teams up to three developers, with sync capabilities for larger groups via a team dashboard.