Sgraal Memory Governance

Name: Sgraal Memory Governance
Author: sgraal-ai

Local

github.com/sgraal-aiMemory & Knowledge0Updated 16w ago

Repository

About Sgraal Memory Governance

Sgraal Memory Governance evaluates agent memory reliability through the Sgraal API and returns a risk score and four-band decision before action. It governs an existing memory store rather than replacing it.

Key features

01Assess memory reliability and score risk
02Return a four-band action decision
03SDKs for multiple frameworks
04Zero-knowledge preflight endpoint

Use cases

01Check memory reliability before an agent acts on it
02Warn or ask the user about risky memories
03Block actions based on unreliable memories

MCP support

The repository includes a TypeScript @sgraal/mcp SDK that wraps Sgraal's preflight API; the input does not specify which tools an MCP server exposes.

#memory #governance #ai-agents #security #preflight #agent-memory #llm-security

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config

{
  "mcpServers": {
    "sdks": {
      "command": "npx",
      "args": [
        "-y",
        "@sgraal/mcp"
      ],
      "env": {
        "SGRAAL_API_KEY": "<SGRAAL_API_KEY>"
      }
    }
  }
}

Run commandnpx -y @sgraal/mcp

Environment variables to set

SGRAAL_API_KEYSecret

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages

npm@sgraal/mcpstdio

Tools

Saved tool list · updated 3w ago

Parameter	Type	Required	Description
memory_state	array	Yes	Memory entries to evaluate. Each entry needs: id, content, type, timestamp_age_days, and optionally source_trust (0-1), source_conflict (0-1), downstream_count.
action_type	enum	No	How critical is the action? informational < reversible < irreversible < destructiveinformational \| reversible \| irreversible \| destructive
domain	enum	No	Domain context — higher-risk domains have stricter thresholdsgeneral \| customer_support \| coding \| legal \| fintech \| medical

Details

From the project README

Sgraal SDKs

Official open-source SDKs for Sgraal — vendor-neutral memory immune system for AI agents.

Each SDK wraps the public POST /v1/preflight endpoint and returns a four-band decision (USE_MEMORY / WARN / ASK_USER / BLOCK) so your agent can act on its memory only when the memory is reliable.

SDK	Language	Install	Status	Source
langchain-sgraal	Python	`pip install langchain-sgraal`	Production	`./langchain-sgraal/`
mem0-sgraal	Python	`pip install mem0-sgraal`	Production	`./mem0-sgraal/`
openai-sgraal	Python	`pip install openai-sgraal`	Production	`./openai-sgraal/`
crewai-sgraal	Python	`pip install crewai-sgraal`	Production	`./crewai-sgraal/`
autogen-sgraal	Python	`pip install autogen-sgraal`	Production	`./autogen-sgraal/`
sgraal-rag	Python	`pip install sgraal-rag`	Production	`./sgraal-rag/`
@sgraal/mcp	TypeScript / Node	`npm install @sgraal/mcp`	Production	`./mcp/`
llamaindex-sgraal	Python	source-available — not yet on PyPI	Beta	`./llamaindex-sgraal/`
haystack-sgraal	Python	source-available — not yet on PyPI	Beta	`./haystack-sgraal/`
semantic-kernel-sgraal	Python	source-available — not yet on PyPI	Beta	`./semantic-kernel-sgraal/`

What is Sgraal?

Sgraal is the immune system for AI agent memory. It evaluates whether your agent's memory state is reliable enough to act on, returning a risk score and a recommended action: proceed, warn, ask the user, or block.

Learn more at sgraal.com:

Tutorial — 5-minute walkthrough including the 3 decision bands
API reference — full endpoint contract
Threat model — what Sgraal catches, doesn't replace, and complements
Zero-Knowledge preflight — /v1/preflight/zk for GDPR / HIPAA / data-residency contexts

Status legend

Production — substantive integration with tested wrapper surface and field-tested patterns.
Beta — thinner integration (typically <50 LOC); functional but a smaller wrapper surface. Re-classified per the project's SDK tier reframe.

License

All SDKs in this repo are licensed under Apache 2.0. See LICENSE.

All SDKs are wrappers for the paid Sgraal API. Calibrated scoring requires a Sgraal account. Get one at sgraal.com.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

Read the getting-started guide

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.