scritty

Scritty sits at an interesting intersection: it's both a terminal emulator and a memory layer for AI-assisted development. For developers who hop between Claude Code, Codex, Copilot, and others, the value is obvious—no more repeating context or losing track of what an agent did in a previous session. The automatic capture at the process level means you don't have to remember to save anything; it just records. The hybrid vector+keyword search (CTRL+SHIFT+M) is fast and lets you find both exact terms and semantic matches, which is more flexible than simple grep. The MCP server integration is a standout: it lets agents pull relevant context from past sessions, so you can tell an agent 'remember our coding conventions' and it actually can. For teams, the per-tenant control plane and admin overrides provide governance over what gets shared, which is important if you're concerned about code context leaving the organization. That said, Scritty is in early access, and that shows. Integrations are limited to a few editors and CLI; there's no public API yet, which limits automation and custom tooling. Cloud sync could raise privacy concerns for some, and no on-premise option means teams with strict data residency rules may be excluded. The free price is compelling, but you should weigh the early-stage risk against the productivity gains. If you're a solo developer or small team using multiple agents, the memory features are likely worth it. If you rely on a single agent with built-in memory, or if you need enterprise-grade deployment, you may want to wait or look elsewhere. Scritty's roadmap—rules engine, pluggable vector backends, offline mode—shows ambition, and the core functionality works as advertised, but it's not yet a mature product. I'd recommend trying it in a non-critical project first, then scaling if it sticks.