Sgraal learns and adapts automatically. Your agents keep running.
Self-improving thresholds, autonomous healing, 6 SDKs, 14 frameworks — zero manual tuning required.
Get API Key arrow_forwardWithout Sgraal
With Sgraal
Systems don't fail all at once. They degrade — until something breaks.
Reinforcement learning records every outcome. Calibrated thresholds improve automatically.
Predictive health forecasts 30 days ahead. Autonomous heal triggers before problems surface.
pip install sgraal · langchain-sgraal · @sgraal/mcp. Zero-config. 14 framework integrations.
RL Q-table: every decision outcome improves thresholds. Meta-learning: cross-agent patterns. Calibrated thresholds per domain, action type, memory type. Weight export/import.
No manual tuning. Every outcome makes future decisions more accurate.
Predictive Health: first_block_day forecast 30 days ahead (validated on internal test suite — accuracy improving with each outcome submitted). Autonomous Immune: auto-heal after 3+ BLOCKs. Autonomous Rollback: wrong email → revoke. Autonomous Pruning: Shapley-weighted removal.
Problems are resolved before you notice them. Your fleet runs itself.
Python SDK: async + sync, Pydantic v2. LangChain: pip install langchain-sgraal. MCP Server: npm install @sgraal/mcp — Claude Desktop, Cursor, Windsurf. Also: CrewAI · AutoGen · LlamaIndex · mem0 · CLI.
No migration. No rewrite. Drop into your existing stack in minutes.
p50 latency: ~1.8s. p99: ~2.5s. Full 83-module scoring pipeline. 166+ API endpoints. Upstash Redis for state management.
Full 83-module scoring on every decision. Worth the 2 seconds for irreversible actions.
Coordinate unlimited agents. Cross-agent memory sharing with RBAC. Fleet-wide health dashboard. Agent lineage tracking. Memory inheritance between agent generations.
Scale to any number of agents with shared, governed memory and zero coordination overhead.
Every outcome trains the RL model. Thresholds improve automatically.
# Install
# pip install sgraal
# npm install @sgraal/mcp
from sgraal import SgraalClient
client = SgraalClient(api_key="sg_demo_playground")
# Preflight with RL-calibrated thresholds
result = client.preflight(
memory_state=[{"id": "mem_001", "content": "...",
"type": "semantic", "timestamp_age_days": 5}],
action_type="irreversible",
domain="fintech"
)
print(result.recommended_action) # USE_MEMORY
print(result.rl_adjustment) # {"rl_adjusted_action": "USE_MEMORY", "confidence": 0.91}
# Report outcome to train the RL model
import requests
requests.post(
"https://api.sgraal.com/v1/outcome",
headers={"Authorization": "Bearer sg_demo_playground"},
json={"outcome_id": result.outcome_id, "status": "success"}
)
Every agent that passes Sgraal's memory governance checks can display a verified badge — cryptographically signed, publicly verifiable, W3C Verifiable Credentials compatible.
Badge integrity verified via HMAC-SHA256. Revocable. Auditable.
Get your badge arrow_forwardConnect Sgraal to your existing workflows without writing code. BLOCK events trigger Zapier automations. WARN thresholds fire Make.com scenarios. Memory governance becomes part of your no-code stack.
Trigger any of 5,000+ Zapier apps when Sgraal blocks a decision or detects an attack.
POST /v1/zapier/webhook
{
"webhook_url": "https://hooks.zapier.com/...",
"trigger": "block"
}
Build visual workflows that respond to memory governance events in real time.
POST /v1/make/webhook
{
"webhook_url": "https://hook.make.com/...",
"trigger": "warn"
}
Works with 5,000+ Zapier apps and Make.com's visual workflow builder.