Neura Market
piia-engram logo

piia-engram

Free

跨工具持久化 AI 记忆——你的身份、偏好、经验教训跟随你在 Claude Code、Cursor、Codex 等工具间流转。本地存储，无需云端。

FreeFree tier
Type
Open Source
Visit Website GitHub (40 stars)

About piia-engram

Piia Engram is an open-source, local-first personal AI identity layer designed for developers using multiple MCP-compatible coding tools such as Claude Code, Codex, Cursor, Windsurf, and others. It stores your preferences, quality standards, lessons learned, decisions, and project context in local files (~/.engram/ as JSON/Markdown) that you own and control. AI agents can read this identity layer to generate context-aware responses, while high-risk writes (credentials, shell commands, MCP config, permission rules) require your explicit approval. The tool includes a Memory Lens (engram preview --html) to inspect exactly what any AI caller receives. It is portable across tools, requires no cloud account, and offers a strict approval mode for maximum governance.

Key Features

Local-first storage: identity and knowledge live under ~/.engram/ as JSON/Markdown
Cross-tool portable: works with Claude Code, Codex, Cursor, Windsurf, and other MCP-compatible tools
User approval system: AI writes locally; high-risk items (credentials, shell commands, MCP config) wait for review; low/medium writes auto-absorbed but auditable
Strict mode (ENGRAM_APPROVAL=strict): gates every write for maximum governance
Memory Lens (engram preview --html): inspect exactly what any AI caller would receive
Open source: install with pip from GitHub, contributions welcome

Pros & Cons

Pros
  • No cloud dependency: all data stored locally, user owns the files
  • Transparent and auditable: all AI writes can be reviewed and reversed
  • Portable across multiple MCP coding tools, reducing context switching
  • Open source with active community (680 commits, 170 stars, 10 forks)
  • Supports strict approval mode for security-conscious users
Cons
  • Requires setup via pip and familiarity with MCP ecosystem
  • Limited to MCP-compatible tools; not useful for standalone AI assistants
  • May add overhead for single-tool users who don't need cross-tool portability
  • High-risk writes require manual approval, adding workflow friction

Best For

Maintain consistent developer identity and preferences across multiple AI coding assistantsPreserve lessons learned and project context to avoid re-explaining to each toolEnable cross-tool continuity: e.g., a memory written by Claude Code read back by CodexProvide a user-owned knowledge layer above native tool-specific memories

FAQ

Why not just use native memory from Claude Code or Cursor?
Native memories are scoped to one tool or workspace. Piia Engram provides a portable identity layer above them, allowing context to follow you across tools while keeping files you own.
How does the user approval system work?
AI writes locally; high-risk items (credentials, shell commands, MCP config, permission rules) wait for your review. Low/medium writes are auto-absorbed but fully auditable and reversible. Setting ENGRAM_APPROVAL=strict gates every write.
Where is my data stored?
All identity and knowledge files are stored locally under ~/.engram/ as JSON and Markdown files. No cloud account is required, and you can inspect or delete the data at any time.