Ghost-Frame/engram
Description: Persistent memory system for AI agents. store, search, and recall context across sessions
Platform-specific configuration:
{
"mcpServers": {
"engram": {
"command": "npx",
"args": [
"-y",
"engram"
]
}
}
}Add the config above to .claude/settings.json under the mcpServers key.
<div align="center">
The cognitive layer for AI agents.
Memory, personality, reasoning, and trust in a single self-hosted system that learns, forgets, and grows.
[](CHANGELOG.md) [](LICENSE) [](https://nodejs.org) [](docker-compose.yml)
</div>
<div align="center">
Quick Start · Features · Architecture · API · Growth · CLI · SDK · MCP · Deploy
</div>
---
---
FSRS-6 spaced repetition with power-law forgetting. Hybrid search fuses vector similarity, full-text matching, personality signals, and graph traversal into a single ranked result. Memories strengthen when accessed and fade when ignored.
Long memories are automatically broken into self-contained atomic facts. Each fact is independently searchable and linked back to its source. Search and recall can filter to facts only (facts_only) or exclude them (exclude_facts). Context assembly groups child facts under their parent for coherent presentation. Decomposition runs on store and can sweep existing memories retroactively via the admin API.
Extracts preferences, values, motivations, decisions, emotions, and identity markers from conversations. Synthesized personality profiles are automatically injected into /recall and /context responses, so every agent interaction is personality-aware without configuration.
Detects contradictions between stored facts. Generates reflections. Derives new knowled
Loading reviews...