Skip to main content
rlcli wraps the official tinker CLI for checkpoint, run, and session so you can manage checkpoints and runs against your local SkyRL server without switching tools. The passthrough commands forward all extra arguments unchanged and pre-configure the Tinker base URL and API key.

How it works

Each passthrough command runs:
  • rlcli checkpoint ... → python -m tinker.cli checkpoint ...
  • rlcli run ... → python -m tinker.cli run ...
  • rlcli session ... → python -m tinker.cli session ...
rlcli sets TINKER_BASE_URL to http://localhost:8000 if unset, and TINKER_API_KEY to tml-dummy if unset. All other flags and positional arguments are forwarded directly to the upstream CLI.

Shared flag

string
default:"http://localhost:8000"
Tinker API server URL. Sent to the underlying tinker CLI.

Common tasks

List checkpoints:
List training runs:
Get details for a specific run:
List active sessions:

Exit codes

The exit code from the passthrough is forwarded directly from the subprocess. If the underlying tinker CLI exits with code 1, rlcli also exits with code 1.

Further reading

For the full flag surface of each subcommand, see the upstream Tinker CLI documentation at https://tinker-docs.thinkingmachines.ai/.