Manage local AI models via Ollama.

The model command provides a focused surface for day-to-day local model operations: list what is installed, pull new models, remove ones you no longer need, update to the latest tag, and benchmark inference speed.

All subcommands talk to the Ollama API. Ollama must be running:

nself plugin install ollama
nself start

nself model list [--json]

Show every model currently downloaded in the local Ollama store.

nself model list

nself model list --json | jq '.[].name'

NAME                                        SIZE  MODIFIED
------------------------------------------------------------------------
gemma-3-4b                               2.5 GB  2026-04-01    [default]
llama3.2:3b                              1.9 GB  2026-04-10

nself model pull <model>

Download a model from the Ollama library. Model names follow the Ollama tag format:

Common models and approximate hardware requirements:

nself model pull gemma-3-4b
nself model pull llama3.2:3b
nself model pull mistral

Pulling llama3.2:3b...
Model llama3.2:3b ready.

nself model remove <model>
nself model rm <model>
nself model delete <model>

Delete a downloaded model from the local Ollama store. The disk space is freed immediately.

nself model remove gemma-3-4b
nself model rm llama3.2:7b

Removed gemma-3-4b.

nself model update <model>

Re-pull a model to pick up the newest weights for its tag. Ollama only downloads layers that have changed, so updates are incremental.

nself model update gemma-3-4b

Updating gemma-3-4b...
Pulling gemma-3-4b...
Model gemma-3-4b ready.

nself model benchmark <model> [--runs N] [--prompt "..."] [--json]

Send a standard prompt to the model N times and report:

The default prompt is: Explain what a Merkle tree is in two sentences.

# Basic benchmark
nself model benchmark gemma-3-4b

# 20-run benchmark for a stable p99
nself model benchmark llama3.2:3b --runs 20

# Custom prompt
nself model benchmark mistral --prompt "Summarize TCP in one sentence."

# JSON output for scripting
nself model benchmark gemma-3-4b --json | jq '.avg_tok_s'

Benchmarking gemma-3-4b  (5 runs)...

  Model:        gemma-3-4b
  Runs:         5 / 5 succeeded
  Avg latency:  1842 ms
  p99 latency:  2103 ms
  Tok/s:        14.2
  Total tokens: 143
  Elapsed:      9.21s

{
  "model":        "gemma-3-4b",
  "runs":         5,
  "avg_tok_s":    14.2,
  "p99_ms":       2103.0,
  "total_tokens": 143,
  "errors":       0
}

• nself ollama, manage the Ollama service container (install, status)
• nself ai local, manage models via the plugin-ai registry
• nself plugin, install and manage ɳSelf plugins
• Home