Skip to content

ollama-cli

Recipe card from the charly-ollama plugin (Images — the deployable catalog).

The compiled-in charly ollama management CLI for a deployed Ollama server — a command:ollama plugin (candy/plugin-ollama). It talks to the server’s HTTP API via the official upstream Go client (github.com/ollama/ollama/api), so it works identically compiled-in or out-of-process and needs no reverse channel.

--server <url> flag > OLLAMA_HOST env > http://127.0.0.1:11434. A schemeless value gets http:// prepended; 0.0.0.0 (the candy’s server-bind value) maps to 127.0.0.1 for client use. The deployment’s actual host port comes from charly status <name> (auto-allocated on 127.0.0.1 at deploy).

  • charly ollama list — pulled models (NAME / SIZE / MODIFIED).
  • charly ollama ps — models loaded in memory (NAME / SIZE / VRAM / UNTIL).
  • charly ollama pull <model> — pull with compact transfer progress on stderr.
  • charly ollama rm <model>... — delete one or more models.
  • charly ollama show <model> — details (format/family/parameters/quantization) + Modelfile.
  • charly ollama cp <src> <dst> — copy a model.
  • charly ollama create <name> --from <base> [--system <text>] — create from a base model. Full Modelfile authoring stays with the upstream CLI (the ollama candy’s host alias).
  • charly ollama push <model> — push with progress.
  • charly ollama run <model> [prompt…] — streaming generate; a simple interactive chat REPL (/bye or Ctrl-D exits) when no prompt is given.
  • charly ollama stop <model> — unload from memory (zero-keepalive generate, the upstream mechanism).
  • charly ollama version — the server version.

Every leaf takes --server <url> (anywhere in the args). A non-zero server response propagates as the command’s own non-zero exit.

  • candy/ollama — deploys the server (supervisord ollama serve on 11434).
  • The candy’s host shell alias: ollama (charly alias install ollama) — execs the upstream CLI INSIDE the container.
  • This plugin — manages the server from the HOST over HTTP.