
Engram
by Gasmanc
Persistent, token-efficient project memory for AI agents using a local SQLite database.
What it does
Engram provides a structured memory system that solves the 'context loss' problem when switching AI sessions or agents. It stores project decisions, task statuses, and file summaries in a local SQLite database, allowing agents to quickly orient themselves with tiered, role-based context.
Tools
get_context: Retrieves filtered project context based on role (build, review, debug, resume) and level (minimal, standard, full).add_decision: Records architectural decisions with rationale and alternatives.update_task: Manages task progression and status updates.summarize_file: Maintains up-to-date summaries of project files with hash tracking.start_session: Initiates a tracked work session.end_session: Records handoff notes for the next agent.search: Performs full-text search across all stored memories.
Installation
First, install the binary:
cargo install --path .
Then add to claude_desktop_config.json:
{
"mcpServers": {
"engram": {
"command": "engram",
"args": ["serve"]
}
}
}
Supported hosts
- Claude Desktop
- Cursor
- Claude Code
Quick install
cargo install --path .Information
- Pricing
- free
- Published
- 4/20/2026






