runtime
Manage the runtimes (Ollama, cloudflared, llama-server) the agent supervises.
ownllm runtime list [--json]
ownllm runtime update [--kind <kind>]
ownllm runtime ensureOwnLLM treats Ollama, cloudflared, and llama-server as runtimes —
versioned dependencies pinned by a manifest the site publishes at
/api/v1/runtimes/manifest. The control plane is the source of truth
for what version each runtime should be on; GitHub Releases is a dev
fallback with a one-hour cache.
list
ownllm runtime list
ownllm runtime list --jsonKIND VERSION STATUS ACCELERATORS
ollama 0.6.4 ready metal, mlx
cloudflared 2025.10 ready -
llama-server - absent -status is one of:
| Status | Meaning |
|---|---|
ready | Installed, version matches the manifest, runtime is reachable. |
outdated | Installed but older than the manifest pin. Run update. |
absent | Not installed. Run ensure. |
degraded | Installed but failing health checks. See logs. |
update
ownllm runtime update # bump every kind to the manifest version
ownllm runtime update --kind ollama
ownllm runtime update --kind cloudflared
ownllm runtime update --kind llama-serverDownloads the manifest-pinned version, verifies the signature, and swaps it in atomically. The current process keeps running until the next supervised restart.
ensure
ownllm runtime ensureA force-install — wipes the local copy, re-downloads from the
manifest, and re-installs. Use it when a runtime is corrupt
(status=degraded or unexpected hashes).
Why a manifest?
Pinning runtimes by version is how we keep the OwnLLM contract
stable across hosts: every paired machine talks to the same
POST /api/show shape, exposes the same tools capabilities, and
behaves predictably under load. Without the manifest, an agent
auto-updating to a breaking Ollama release would drop tool calling
for the whole tenant overnight.
The pin schema is documented in
01_TECHNICAL_SPEC.md
in the repo.