OpenAI Anthropic llama.cpp vLLM OpenRouter Hermes OpenClaw OpenCode MimoCode JCode ZCode NVIDIA Salad Vast.ai RunPod Hostinger DigitalOcean Docker Kubernetes Ubuntu Kali NixOS Kodachi
Capabilities

300+ Features

Every subsystem is independently deployable yet fully integrated into a single operational platform. Self-hosted, auditable, compliant.

300+
Features
22
Agents
21+
Providers
55+
Skills
40+
MCP Servers
Model Router

Intelligent Routing

Classify, route, fallback, cache — every prompt goes through a confidence-scored decision pipeline.

⚙

Task-Aware Classification

Confidence-scored routing per prompt type: coder, reasoning, general, vision.

  • 21+ provider bridges
  • Auto-fallback chains
  • Per-model TTL caching
🔄

Degenerate Output Detection

Catches repetition loops and low-confidence responses before wasting tokens.

  • Perplexity scoring
  • Repetition penalty
  • Auto-retry with different model
🔗

LRU Response Cache

Identical prompts return cached responses with X-Cache: HIT/MISS headers.

  • Per-model cache configuration
  • Automatic invalidation on config change
  • Sub-second cache lookups
Workflow Engine

Visual Pipeline Builder

Design complex multi-step workflows with parallel and sequential execution.

🚀

Parallel Execution

Run multiple steps simultaneously with configurable concurrency limits.

  • Shared context between steps
  • Timeout and retry policies
📋

Validation & Audit

Every step is validated and logged. Full audit trail with timestamps.

  • Step-by-step status tracking
  • JSONL audit log export
📋

Template Library

Pre-built workflows for common patterns: CI/CD, security scan, model training.

  • Import/export workflows
  • Community-contributed templates
Security Suite

Built-In Defense

Self-hosting means you control your attack surface. Every default is secure.

🔒

HITL Approval Gates

6 danger categories require human approval before autonomous execution.

  • Exploit, persistence, credential dump
  • Sandbox traversal, mass operations
⚡

Circuit Breakers

Sliding-window failure detection. Automatic fallback when error ratio exceeds threshold.

  • Configurable window size and ratio
  • Per-backend failure tracking
🔓

Aikido Integration

SAST/DAST scanning from the dashboard. Auto-patch for critical/high vulnerabilities.

  • 25+ security skills included
  • API key rotation (10 keys per provider)

Full feature list

Browse the complete capabilities in our documentation.

Read Docs View Agents