Run one generation on a local Ollama model and return its output. Use for summarization, extraction, reformatting and other mechanical work you do not want to spend your own context on. Supports structured output (format), multi-turn messages, tool definitions, and server-side file reading via files/file_globs so file contents never enter your context. Select the model with model: a literal name, role:NAME, or caps:a+b. Note that reasoning tokens and answer tokens share one num_predict budget — thinking is off by default for that reason. Every response ends with a metrics line naming the model actually used and how it was resolved.

Run many generations in one call. Items are grouped by resolved model and the groups run sequentially, so a cold model load is paid at most once per model instead of thrashing VRAM. Results are returned in input order regardless of execution order, and one failing item never voids the run. Prefer this over many separate ollama_dispatch calls: it is faster and costs you far less context.

List the models installed on the Ollama host with their capabilities, context window, size and whether they are currently loaded. Filter with capability or resident_only; pass refresh after pulling a new model. Pass explain_selector to dry-run the model resolver — it reports exactly which model a selector would run and the full fallback chain, without spending a generation. Start there whenever routing surprises you.

Manage model residency. status lists loaded models with their VRAM use and time until unload. warm pre-loads a model so a following dispatch skips the cold-load cost (which can be ~12s for a large model) — useful before a batch. unload frees the VRAM immediately, which matters because a large model can hold tens of GB and starve everything else on the machine.