Skip to main content

​
Get started in 30 seconds

Prerequisites

​
Install OnCall CLI

npm install -g oncall-cli

​
Start using OnCall CLI

cd your-project

# One‑time setup per project
oncall init -m "Describe this service" -id "<project-id>"

# One‑time setup per machine
oncall login <your-api-key>

# Start OnCall Cluster
oncall cluster

# Start your app under OnCall’s debugger
oncall npm run dev
On first use, oncall login stores your API key in ~/.oncall/config. After that, you can prefix any command with oncall to launch the interactive logs + AI chat interface.

​
What it does for you

  • Real‑time debugging: run any command through OnCall CLI, stream logs live, and ask the AI to explain failures or propose fixes while the process runs.
  • Multi‑service log Q&A: query logs across multiple services that share the same project ID (frontend, backend, workers, microservices).
  • Contextual code inspection: let the AI read targeted slices of code (when code_available: true) to explain or fix issues it detects in logs.
  • Cross‑service correlation: with the local cluster server, AI can reason across all registered services in a project, pick which service to inspect, and trace issues end to end.
  • Tooling you control: AI uses local tools (read_file, grep_search, tail_logs, grep_logs, get_recent_errors, read_logs) and respects logs_available / code_available flags.

​
Why devs love OnCall CLI

  • Works in the terminal you already use (Ink‑based UI with split panes for logs and AI chat; keyboard shortcuts for focus, view toggles, clearing panes).
  • Live log awareness: logs are captured character‑by‑character via node-pty, so AI always has the freshest context without reloading.
  • Flexible access control: per‑service oncall.yaml flags gate code and log access; safe for production (log‑only) or full‑access development.
  • Multi‑service ready: cluster mode groups services by project ID, enabling unified debugging across devices or pods with minimal setup.
  • Concrete, actionable help: AI doesn’t just diagnose; it reads code (when allowed), searches it, inspects logs, and returns specific fixes or steps.