Persistent Memory Operating System for AI Agents
Provide your autonomous agents with biological-inspired memory structures: Episodic events, semantic facts, and procedural skills. Securely audited with cryptographic integrity and structured with natural decay parameters.
Interactive Memory Playground
Interact with the client-side simulator to see how Kyros parses, secures, decays, and propagates state in real time. Or view the full sandbox on the Simulation Page.
User mentioned their name is Alice and they are building a fintech application.
User prefers Python for backend services and PostgreSQL for data storage.
System Specifications & Features
Three Memory Modules
Episodic (conversations), semantic (structured facts), and procedural (predefined workflows) subsystems integrated within a unified architecture.
Ebbinghaus Temporal Decay
Memories fade dynamically according to category decay rates. Prevents token bloat, keeps search queries relevant, and optimizes context windows.
Cryptographic Tamper Auditing
Protects memory chains using SHA-256 hashes and Merkle tree verification logic. Instantly detects external injections and poisoning attacks.
Causal Relationship Chains
Establishes explicit parent-child links between independent memory nodes, allowing the agent to audit reasoning chains and trace causality.
Adaptive Belief Propagation
Graph-based conflict resolution updating network confidence states iteratively using breadth-first traversal whenever new info is parsed.
Zero-Code API Proxy Mode
Point an existing LLM wrapper base_url directly to the Kyros proxy server. Intercepts request payloads to inject and log context automatically.
Up in 60 Seconds
Self-host the memory server using Docker or deploy the Python and TypeScript SDK wrapper.
# Install dependencies
pip install kyros-sdk
# Initialize client
import kyros
client = kyros.Client(
base_url="http://localhost:8000",
api_key="mk_live_default_dev_key_123456"
)
# Store episodic memory
client.remember("agent-1", "User prefers strict typing.")
# Recall memories
results = client.recall("agent-1", "What type guidelines does user follow?")
print(results.results[0].content)
# -> "User prefers strict typing."