TrylleDocs
CLI

Trylle CLI

Use the try command for repositories, pull requests, issues, CI, AI helpers, and agent workflows.

The Trylle CLI is the terminal interface to the Trylle platform. Use try for platform resources and git or jj for local version-control operations.

npm install -g @trylle/trylle-cli
try auth login
try whoami

Two executable names

The npm and PyPI packages install both try and trylle-cli. This documentation uses the shorter try name. A Cargo install provides only trylle-cli.

Choose a guide

Understand the execution boundary

The CLI combines several kinds of commands. Their credentials and side effects are deliberately separate.

SurfaceExamplesWhere it runsCredential
Trylle platformrepo, pr, issue, ciPublic Trylle APITrylle CLI Key or bot token
Local VCSclone, parts of stack, local diffSystem Git or local jj repositoryGit transport credential when needed
Direct AIask, explain, draftConfigured AI providerProvider API key, or none for local Ollama
Local agentagentInstalled Codex, Claude, or OpenCode CLIThe local agent's own authentication
Diff TUIdiffLocal terminal; Trylle API for PR inputNone for local diffs; Trylle credential for PRs

Direct AI calls are bring-your-own-key requests to the selected provider. They do not use a Trylle CLI Key and are not charged against Trylle-hosted AI usage.

Resolve repository context

Inside a checkout, commands that accept -R/--repo infer OWNER/NAME from a recognized Git remote. Inspect what the CLI resolved before a write:

try repo context

Outside a checkout, or when multiple remotes make the target ambiguous, pass the repository explicitly:

try pr list -R OWNER/NAME
try issue view 42 -R OWNER/NAME

Commands such as try repo view, try clone, and try repo clone accept OWNER/NAME positionally instead.

Human and structured output

Most platform commands default to formatted terminal output:

try status

Scripts and agents can request structured output where the leaf command supports it. try explain, try draft, and try diff also offer structured output. try ask and try agent do not because they are interactive or streaming workflows.

A reliable operating pattern

  1. Inspect authentication with try auth status and try whoami.
  2. Resolve the target with try repo context or pass -R OWNER/NAME.
  3. Run try <group> <command> --help before a consequential write.
  4. Execute the narrow command and inspect its result.
  5. Re-read the changed resource, then use try browse ... --no-open to return its URL.

The command reference lists every top-level group and the caveats that matter in automation.

On this page