Visor on GitHub

Project Overview: Visor came out of running AI-powered code review in practice and hitting the same wall every team hits: the AI part is easy, but wiring it into real workflows means writing endless integration glue — one script for CI, another for chat, another for the CLI, all subtly different.

Visor replaces that glue with declarative YAML pipelines. You define steps — AI prompts, shell commands, MCP tools, HTTP calls — declare dependencies between them, and Visor builds the execution DAG, parallelizes what it can, and routes failures. The same configuration then runs everywhere: as a GitHub Action reviewing pull requests, as a CLI, as a Slack or Telegram bot, or as an HTTP server — eight runtime modes in total.

It is AI-provider agnostic (Anthropic Claude, Google Gemini, OpenAI, AWS Bedrock), supports composable assistant workflows with skill-based intent routing, and — because pipelines are code — ships with a YAML-native testing framework with mocks and fixtures, plus OpenTelemetry observability for debugging runs in production. Written in TypeScript, it’s the orchestration layer of the Probe Labs ecosystem, sitting on top of Probe’s code intelligence.

Outcome

Visor powers zero-boilerplate code review automation and multi-step AI workflows for teams who want their automation versioned, tested, and portable across interfaces. Together with Probe and Proof, it completes a stack where AI agents can understand code, act on it, and have their output verified.