Dynamic Planning
State a goal in plain language. The LLM planner selects agents, attaches MCP tools, and builds a multi-step execution plan automatically.
182 Agent Templates
YAML catalog spanning OpenAI, Anthropic, Ollama, Hugging Face, vLLM, and JetStream. Hardware-aware routing filters by CPU/GPU/TPU and VRAM.
7 MCP Integrations
Brave Search, Tavily, Exa, Home Assistant, URL fetch, filesystem, and memory knowledge graph — credential-gated and Kubernetes-ready.
Agent Skills
YAML procedural playbooks — release checklists, PR review, domain how-tos — injected into tasks or backstory; composes with MCP tools.
Platform Agent Harness
Tiered per-catalog verification (L0–L3): static YAML checks, connectivity probes, smoke tasks, and capability rubrics. L0/L1 run in CI on every PR.
Three Execution Backends
In-process CrewAI (default), subprocess workers, or Kubernetes Jobs with a shared PVC run store. Same YAML catalogs across all three.
Sessions, Learning & KB
Planner history persists across runs. SQLite full-text knowledge base injects relevant snippets into future plans. Learning loop tracks per-provider success rates.
Domain Verticals
Healthcare and Logistics overlays ship out of the box. Each adds domain context, specialist agent YAML, and a one-command web UI entry point.