scritty

About scritty

scritty is a terminal emulator that records conversations from command-line AI coding agents like Claude, Codex, and Copilot. It indexes these exchanges into a local, searchable vector store and serves the data back to agents via the Model Context Protocol (MCP) or to users via the command line.

Review

Developers frequently switch between different AI coding agents and lose conversational context in the process. scritty addresses this by acting as the underlying terminal environment, capturing every exchange regardless of the active provider. The tool then makes this accumulated history queryable for subsequent sessions.

Key Features

• Captures terminal output directly from the running process, tagging each exchange with the specific provider without relying on vendor APIs.
• Indexes captured text using an embedded vector store by default, with options to connect external databases like Qdrant, pgvector, Chroma, or Weaviate.
• Reads a prompt.toml configuration file to inject custom rules into every message before it reaches the active agent.
• Synchronizes the active terminal session across desktop, browser, and mobile devices through a progressive web application.

Pricing and Value

The tool operates on a local-first, paid model. A personal subscription costs $19.99 per month and includes a 14-day pilot period. There is no permanent free tier. The pricing reflects the local processing and storage requirements, while team features like SSO, SAML, and audit logs scale the platform for organizational use.

Pros

• Maintains a single searchable corpus across multiple AI providers, preventing context loss when switching tools.
• Classifies captured content into normal, thinking, and tool-call spans at the time of recording to improve retrieval accuracy.
• Keeps all data and processing entirely on the user's local machine without requiring a cloud account.
• Allows agents to query their own past turns and the past turns of other agents through the MCP server.

Cons

• Stores raw terminal output as plaintext in the local index, meaning API keys or environment variables are not automatically redacted at the time of capture.
• Lacks a hard delete mechanism for incorrect or superseded memories, relying instead on relevance decay and noise marking to demote bad data.
• It is not well suited for regulated environments that mandate automated secret redaction at the exact moment of data capture, as this feature is currently planned for the roadmap rather than shipped.

scritty fits into the workflow of developers who regularly rotate between different command-line AI agents and need to preserve debugging context. It serves individual programmers who want local control over their AI interaction history. The tool is less appropriate for teams that need immediate, automated secret redaction without waiting for future updates.

Open 'scritty' Website