Reference

satyrn commands

Without the installed command, write python3.14 satyrn.py COMMAND in the project folder (Windows: py -3.14 satyrn.py COMMAND).

Command	What it does
(none)	Show what is ready and offer each missing step: tools, model, server.
install	Install the satyrn command (linked to this folder) and tab completion for your shell. Run as python3.14 satyrn.py install.
uninstall	Remove the command and the tab completion. The folder stays.
code [DIR] [-p REQUEST] [-m MODEL] [--pi | --open-interpreter]	Work with the model in DIR (default: the current folder): a conversation, or one request with -p. MODEL is a listed model, by name or number; --pi or --open-interpreter overrides your default coding agent once. A conversation starts with a background warm-up (Speed); --no-warmup turns it off.
models	Rate every prepared model for this computer, with benchmark results.
use MODEL	Make a listed model your default, by name or by its number in satyrn models.
setup [--harness NAME] [--model M] [--no-model]	Download and verify Ollama, Open Interpreter, Node.js and pi, write their settings, download the default model (or M). --harness pi or --harness open-interpreter sets the default coding agent.
pull MODEL	Download a model; asks first if the available build differs from pins.toml.
start / stop	Start the Ollama server in the background and load your default model / stop it.
serve	Run the server in the foreground; Ctrl+C stops it.
status	Show the server, the installed models and the loaded ones.
check [--harness NAME] [--model M] [--effort E] [--timeout S]	The model writes sqrt.py; the program is then run and its output checked.
bench [MODEL…] [--harness NAME] [--runs N] [--warm] [--instructions FILE] [--set KEY=VALUE] [--effort E] [--timeout S]	Time models on the same task with a coding agent (default: yours); default: every model in config/models.toml, one run each.
verify	Re-check every pinned archive, unpacked tool and downloaded model (accepted model versions pass).
pi ARGS…	Run the project’s pi with the project’s settings; all arguments are passed on.
interpreter ARGS…	Run the project’s Open Interpreter with the project’s settings; all arguments are passed on.
ollama ARGS…	Run the project’s Ollama command line, for example satyrn ollama list.
completion SHELL	Print the tab-completion script for bash, zsh, fish or powershell.

--effort sets the reasoning effort for models that have one (gpt-oss:20b: low, medium, high). --timeout is the time limit per run in seconds, 3600 by default.

make targets

The Makefile is only a shortcut on systems with make; all logic is in satyrn.py.

Target	Same as
make setup, start, stop, serve, status, verify, check, bench	satyrn …
make oss	satyrn code
make test, test-live, lint, format, html, livehtml	see Contributing
make install	python3.14 satyrn.py install
make install-codegen	install satyrn-codegen with pipx (Satyrn’s own engine)

Files you may edit

pins.toml

Default model, model digests, context window, port, tool versions and checksums.

config/context/pi.md, config/context/open-interpreter.md

The context (system prompt) each coding agent gives the model.

config/models.toml

The models Open Interpreter knows how to drive: reasoning efforts, sizes and benchmark results for satyrn models.

config/interpreter.toml

Open Interpreter settings. After editing, run satyrn setup again.

Environment

satyrn.py sets these for the tools it starts; you do not need to set them yourself.

Variable	Value	Why
INTERPRETER_HOME	.runtime/interpreter-home	keeps the interpreter’s settings out of ~/.openinterpreter
CODEX_OSS_BASE_URL	http://127.0.0.1:11434/v1	where --oss finds the model server
OLLAMA_HOST	127.0.0.1:11434	the server listens on this computer only
OPEN_INTERPRETER_PRODUCT_NAME	Satyrn	the name Open Interpreter shows; not supported by 0.0.60 yet
PI_CODING_AGENT_DIR	.runtime/pi-home	keeps pi’s settings, model list and sessions out of ~/.pi
PI_TELEMETRY, PI_SKIP_VERSION_CHECK	0, 1	no install telemetry and no update checks
OLLAMA_MODELS	.runtime/models	model files stay in the project
OLLAMA_CONTEXT_LENGTH	32768	context window offered to the interpreter
OLLAMA_KEEP_ALIVE	30m unless you set it	how long an idle model stays in memory
HOME, USERPROFILE (Ollama only)	.runtime/home	Ollama’s key pair stays out of ~/.ollama