Protects LangChain agents from prompt injection by intercepting malicious prompts or SQL attempts in kernel space and calculating the blast radius.
Drop-in OpenAI-compatible proxy with bidirectional guardrails, blocking ingress prompt injections and redacting sensitive egress data such as Turkish IDs, credit cards, IBANs, and API keys.
Provides kernel-level interception of destructive SQL queries and can synthesize autonomous PostgreSQL ROLLBACK statements with causal forensics after an attack.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@agent-ebpf-mcp Show me current eBPF security policies and their statuses"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Enterprise Documentation • Architecture Whitepaper • Security Manifest • Contact Sovereign Sales
As enterprise deployments shift from stateless conversational LLMs to stateful, autonomous execution swarms (FastMCP, LangGraph, AutoGen), traditional application-layer guardrails fail to mitigate critical attack vectors. Semantic evaluators and regex interceptors introduce prohibitive latency penalties (500ms–2,500ms) and remain fundamentally susceptible to Time-of-Check to Time-of-Use (TOCTOU) exploitation, token obfuscation, and raw socket data exfiltration.
KSEC relocates defense boundary verification from user-space down to the Linux eBPF LSM & XDP kernel substrate. By coupling sub-35μs cryptographic intent-lease verification with in-kernel packet inspection, KSEC provides mathematical finality against sandbox escapes before execution passes the libc boundary.
+-------------------------------------------------------------+
| Enterprise AI Swarm / LLM Orchestrator |
| (Claude FastMCP, LangGraph, Multi-Agent Mesh) |
+-------------------------------------------------------------+
|
[ Cryptographic Intent Lease ]
v
+================================ USER SPACE ============================================+
| KSEC FastMCP Daemon (Zero-Copy Shared Ring Buffer / Memory Lock Verification) |
+========================================================================================+
|
Syscall Execution (`connect`, `bprm_execve`, `write`)
v
+============================ LINUX KERNEL SPACE (Ring-0) ================================+
| |
| +--------------------------+ +-------------------------+ +---------------------+ |
| | eBPF LSM Hooks | | Zero-TOCTOU Engine | | eXpress Data | |
| | security_socket_connect | | Intent-Hash Match | | Path (XDP) | |
| | bprm_check_security | | against Process Lease | | Drop Unverified | |
| +--------------------------+ +-------------------------+ | Egress Packets | |
| \ | +---------------------+ |
| +---------------------------+ / |
| | / |
| v / |
| [ In-Kernel Policy Evaluator (<35μs SLA) ] <------+ |
| | |
| +----------------------+-----------------------+ |
| | | |
| v v |
| [ VERDICT: ALLOW (0-Overhead) ] [ VERDICT: DROP & AUDIT ] |
+========================================================================================+
| |
v v
Physical NIC / Bare-Metal SIGKILL / Audit Vault EventEvaluation Vector | Application-Layer Guardrails (LLM/Regex) | Container Isolation (gVisor/Docker) | KSEC Sovereign Substrate (Ring-0 eBPF) |
Interception Latency |
|
|
|
TOCTOU Exploit Resistance | ❌ Vulnerable (In-memory payload swap) | ⚠️ Partial (Namespace boundary only) | ✓ Absolute (Cryptographic intent-lease) |
Raw Socket Exfiltration | ❌ Unobserved (Layer 7 only) | ⚠️ Coarse network policies | ✓ Granular (Kernel socket inspection) |
Host Resource Overhead | Severe CPU/RAM penalties | Hypervisor virtualization drag |
|
Compliance Auditability | Probabilistic text outputs | System syslog traces | Cryptographically signed kernel audit logs |
curl -sSL https://get.ksec.space | sudo bash# Validate kernel probe attachments and cgroupv2 isolation boundaries
ksec telemetry verify --strict --output=json# 💻 Windows
start.bat
# 🐧 Linux / 🍎 macOS
./start.sh
# 🐳 Docker Container (Kernel Hardened)
docker run -d -p 8000:8000 --privileged -v /sys/kernel/debug:/sys/kernel/debug:ro ghcr.io/ourobx/agent-ebpf:latestKSEC exposes native Model Context Protocol (MCP) endpoints over Server-Sent Events (SSE) and WebSocket streams:
SSE Stream Endpoint: https://ksec.space/sse (or local http://localhost:8000/sse)
JSON-RPC Messages: http://localhost:8000/messages
OAuth 2.0 Auth Discovery: http://localhost:8000/.well-known/oauth-authorization-server
get_security_status: Real-time BPF hook status, memory maps, and blocked syscall metrics.
get_ebpf_status: Ring buffer throughput, dropped packet counters, and probe attachment telemetry.
add_security_rule: Dynamically loads eBPF LSM filtering rules into kernel maps in <10µs.
simulate_query_check: Validates destructive AST statements against declarative kernel policies.
ksec_guard_inspect: Multi-layer token inspection with automated KVKK/GDPR redaction.
For Enterprise Multi-Tenant Governance, Dedicated Bare-Metal Telemetry Nodes, and S3/Vault SOC-2 Archival:
Helm Ingress Chart:
helm install ksec-shield ./deploy/helm/ksec-shield --set existingSecret=ksec-vault-secretProof-of-Hack Simulation: Run the LangChain prompt injection and Ring-0 kernel rollback demonstration:
python demo/terminal_attack_demo.pyDistributed under the Apache-2.0 License. Sovereign Defense Substrate by KSEC Engineering (ourobx).
This server cannot be deployed
Security gateway for AI agents: policy, approval, and audited execution, no secrets shared.
Operate Linux, macOS and Windows from your LLM. Every action runs through an auditable allowlist.
Fail-closed policy guardrails for AI agents running kubectl, terraform, helm, and argocd.
Runtime permission, approval, and audit layer for AI agent tool execution.
