ollama

Version	2026.229.1218
Repo	box/github.com/opencharly/pod-ollama:v2026.237.457

GPU-agnostic Ollama LLM inference server on port 11434 Installs the Ollama binary and runs ollama serve as a custom supervisord service that listens on 0.0.0.0:11434 (no systemd). Two install paths, one resulting layout (/usr/bin/ollama plus the backend libraries under /usr/lib/ollama): on a distro that PACKAGES ollama the package is installed and the package manager owns the version; everywhere else the upstream release tarball is extracted to /usr at the version the var OLLAMA_VERSION selects. That var is PINNED to a release tag and must stay one: the download cache is content-addressed by the sha256 of the URL, so a latest URL would hash to one key forever and serve the first tarball it ever fetched, long after upstream moved on. The candy declares NO GPU dependency — a GPU backend is an IMAGE-level composition choice (compose ollama-cuda or ollama-rocm). What that buys differs by install path, and the difference is worth stating rather than averaging: on a PACKAGED distro the base package is genuinely CPU-only at 66 MiB and a backend is strictly opt-in, which is what keeps this layer small; on the TARBALL path the single upstream archive already contains the CUDA backend, so a tarball consumer pays for it whether or not any box composes ollama-cuda. Pulled models persist under the ~/.ollama models volume. The running service exposes the Ollama HTTP API (e.g. /api/tags) so any composing box can serve and query local models. The companion plugin candy plugin-ollama adds the charly ollama management CLI against a deployed server.

Services

• ollama

Acceptance plan

This candy’s plan: — the runnable spec charly check executes against a live deployment. check: steps are idempotent probes; run: steps change state.

Intent	Step
run	install the upstream ollama release tarball on distros that ship no ollama package (var OLLAMA_VERSION)
check	the ollama binary is installed at /usr/bin/ollama
check	the running ollama service answers the model-tags API with HTTP 200 on port 11434
check	the ollama CLI reports an “ollama version” string
check	the ollama CLI lists models against the live service with exit 0
check	charly ollama help exits 0 and prints the plugin usage, proving the compiled-in command:ollama plugin registers and dispatches
check	charly ollama version reports the server version over the published port, proving a live API round-trip against the deployed server
check	charly ollama list prints the model table header over the published port, proving the list API path against the deployed server
agent-check	models pulled via ollama pull are stored under the ~/.ollama models volume and survive a service restart