Sgraal Memory Governance
About
Deterministic pre-action verdict gate for agent memory — USE/WARN/ASK/BLOCK before an agent acts on a stored memory. Governs your existing store, doesn't replace it.
Details
- Author
- sgraal-ai
- Downloads
- 295
- Categories
- Knowledge Base, AI, Security
Jump to
- Pre-action verdict on every memory read: USE / WARN / ASK / BLOCK
- Blocks unsafe memory before an agent acts on it
- Four deterministic detection layers: timestamp, identity, consensus, provenance
- Multi‑tenant containment limits poisoned memory propagation
- Tested against an internal adversarial corpus of 1,190+ cases
Setting up with Highlight
This MCP is not yet compatible with Highlight’s one-click setup. However, you can still use it with Highlight by following these steps:
- Download and install Highlight from highlightai.com/download
- Navigate to the plugins tab and select "Add Custom Plugin"
-
Configure the plugin with the settings below
Plugin Name
Sgraal Memory GovernanceCommand (node, npx, python, etc.)Please refer to the README for specific instructions on how to obtain API keys or other required environment variables.
- Enable "Start Automatically" if you want the plugin to start when Highlight launches
From the repository
A demo key is provided: sg_demo_playground — no signup needed. The README does not specify further install, configure, or invoke steps.
sgraal_preflight
Check if AI agent memory is reliable enough to act on. Call this BEFORE making any decision based on stored memory. Returns USE_MEMORY (safe), WARN (proceed with caution), or BLOCK (do not proceed).
Claude Desktop / Cursor
Paste into your MCP client config file to install this server.
{
"mcpServers": {
"sgraal memory governance": {
"sgraal": {
"command": "npx",
"args": [
"@sgraal/mcp"
],
"env": {
"SGRAAL_API_KEY": "sg_demo_playground"
}
}
}
}
}McpServers
{
"sgraal": {
"command": "npx",
"args": [
"@sgraal/mcp"
],
"env": {
"SGRAAL_API_KEY": "sg_demo_playground"
}
}
}memory → preflight → action
Deterministic preflight gate for AI-agent memory. Before an agent acts on a stored
memory, Sgraal returns a USE / WARN / ASK / BLOCK verdict — catching poisoned or
stale memory at the read boundary. Deterministic and non-LLM: identical input →
identical verdict. Governs whatever memory store you already use; it isn't another store.
- Pre-action verdict on every memory read: USE / WARN / ASK / BLOCK
- Block unsafe memory before an agent acts on it
- 4 deterministic detection layers: timestamp, identity, consensus, provenance
- Multi-tenant containment: limits how far a poisoned memory can propagate
Evaluation: tested against an internal adversarial corpus (1,190+ cases);
independent third-party evaluation in progress.
Demo key: sg_demo_playground — no signup needed
Sign in to leave a review
Use Google, GitHub, or an email account so ratings stay tied to real people.
No reviews posted yet.
