Radian manual

Radian is a coding agent in one binary. You drive it from a full-screen terminal interface or from one-shot shell commands. Every session is stored as one file, so you can resume, fork, and export your work.

You are here: this page maps the manual. Each line names one page and what it covers.

Start#

  • quickstart: install radian, connect a model, and finish your first task.
  • providers: store keys, choose a provider and model, and control request options.

Use#

  • commands: every slash command and ! shell escape in the full-screen interface.
  • keybindings: every key in the full-screen interface, by area, and how to rebind it.
  • sessions: resume, list, search, fork, export, and attach to sessions.
  • subagent-and-jobs: helper sessions and background shell jobs.
  • permissions: permission tiers, trust, approval, and sandbox behavior.
  • troubleshooting: fix setup, network, and session problems.

Configure#

  • configuration: every key the configuration file accepts.
  • environment: every environment variable radian reads, and the files that change behavior.
  • footer: every footer item, its parameters and placeholders, and the preset contents.
  • themes: switch color themes and import themes from files.

Extend#

  • extensions: which extension mechanism fits which need, where each lives, and a first example.
  • context-files: which project instruction files (AGENTS.md and others) load, from where, in which order, and their limits.
  • hooks: run your own commands when a tool runs or the model stops.
  • mcp: add MCP tool servers that run as local processes over standard input and output.
  • skills: find reusable prompt packages, read them, and run them.
  • memory: save facts across sessions and search them by keyword.

Reference#

  • cli: every command, subcommand, flag, and exit code.
  • pi-migration: draft a Radian configuration from a Pi agent directory.