Quick start
Service + Gateway running with a first gated pipeline in about ten minutes. Get started →
VectorStep Documentation
VectorStep is a webhook-triggered, YAML-configured AI pipeline orchestration service. It receives webhooks from any source (Alertmanager, Grafana, Atlassian, generic JSON), normalises the payload, resolves a named pipeline, and executes a multi-step AI pipeline using pluggable executor backends — with every autonomous action gated by a trust vector built from independent signals.
Tutorials
New to this? Start here after the quick start — build a real agent from scratch, wire it to MCP servers, then turn on gating one signal at a time. Start the tutorials →
Pipeline authoring
The YAML schema: steps, verifiers, grounding, parallel groups, fan-out, flow control, human-in-the-loop. Author pipelines →
Reference
REST APIs, the LLMOutput contract, config file references, MCP servers. Look it up →
The ecosystem
Section titled “The ecosystem”| Component | Role | Distributed as |
|---|---|---|
| VectorStep | The orchestration service: webhook intake, pipeline runner, trust gating, UI, analytics | Container image |
| VectorStep Gateway | WebSocket gateway that runs agents: LLM providers, MCP tools, the full agentic loop | Container image |
| vectorstep-service-mcp | MCP server exposing pipeline authoring, run inspection and analytics to Claude Code/Desktop | PyPI package |
| vectorstep-gateway-mcp | MCP server for authoring and inspecting Gateway agents | PyPI package |