OLLAMA COMMANDS
/ollama SUBCOMMAND [ARGS…]
The /ollama command controls the local Ollama service and manages models from inside the Harvey REPL. Every subcommand maps directly to an ollama CLI operation.
Service control:
/ollama start [debug] Launch ollama serve in the background. If Ollama is already running, prints a warning and exits. Pass debug to also set OLLAMA_DEBUG=1 in the Ollama process; output is captured to agents/logs/ollama-TIMESTAMP.log. Note: OLLAMA_DEBUG is inherited from Harvey’s process — start Harvey with –debug for full diagnostic coverage.
/ollama stop Print a reminder to stop Ollama via your system’s service manager (e.g. systemctl stop ollama). Harvey does not stop the daemon itself.
/ollama status Check whether the Ollama daemon is reachable at the configured URL.
/ollama logs Tail the Ollama service log. Tries ollama logs first, falls back to journalctl -u ollama.
/ollama env Show the Ollama environment variables (OLLAMA_HOST, etc.) as seen by the Harvey process.
Model management:
/ollama list List all installed models. The model currently in use is marked with *.
/ollama ps Show which models are loaded in memory (delegates to ollama ps).
/ollama pull MODEL Download a model from the Ollama registry (e.g. /ollama pull mistral).
/ollama push MODEL Upload a model to the Ollama registry.
/ollama show MODEL Display a model’s Modelfile, parameters, and template.
/ollama create NAME [-f MODELFILE] Create a new model from a Modelfile. Passes all arguments directly to ollama create.
/ollama cp SOURCE DEST Copy an installed model to a new name.
/ollama rm MODEL [MODEL…] Remove one or more installed models.
/ollama use MODEL Switch the active model to MODEL for the current session without restarting Harvey.
/ollama run MODEL [PROMPT] Launch an interactive ollama run session inside the terminal. stdin, stdout, and stderr are passed through directly.
Capability probing:
/ollama probe [MODEL] Run a thorough probe on MODEL (or on all not-yet-probed models when MODEL is omitted). Detects tool-calling support, embedding capability, and whether the model reliably emits path-tagged code blocks (the format Harvey’s auto-execute relies on). Results are cached in harvey/model_cache.db so /ollama list can display them immediately.
/ollama probe-all Re-probe every model currently installed on the local Ollama server, refreshing cached capability data. Useful after pulling several new models or when moving between machines with different model sets. Equivalent to /ollama probe –all.
Model aliases:
/ollama alias NAME FULLNAME Create a short alias for a long model name.
/ollama alias list List all defined model aliases.
/ollama alias remove NAME Remove an alias.