jschoubben
29fe7c5bd0
Model access answered by a NODE, not a licence. `ollama` runs the model server on host network (0.0.0.0:11434) and provides model-access at node scope, serving its port and model — it mints nothing, so it is server-only, no runtime. The `local-model-consumer` requires model-access, gets the endpoint (no secret), and its templated openai.env carries OPENAI_BASE_URL=http://<at>:<port>/v1 + OPENAI_MODEL. One provision, two answers: the mesh's own model behind the same interface as a vendor's. Proven end to end by the mesh-lab local-model bed (green). Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
49 lines
954 B
JSON
49 lines
954 B
JSON
{
|
|
"module": "ollama",
|
|
"version": "1",
|
|
"capabilities": [
|
|
"container-runtime"
|
|
],
|
|
"provides": [
|
|
{
|
|
"name": "model-access",
|
|
"scope": "node"
|
|
}
|
|
],
|
|
"listens": [
|
|
{
|
|
"port": 11434,
|
|
"protocol": "tcp",
|
|
"from": "machine",
|
|
"why": "consumers on this machine reaching the local model server's OpenAI-compatible API"
|
|
}
|
|
],
|
|
"serves": {
|
|
"model-access": {
|
|
"port": 11434,
|
|
"model": "llama3.2"
|
|
}
|
|
},
|
|
"resources": [
|
|
{
|
|
"id": "state",
|
|
"type": "directory",
|
|
"path": "/services/ollama",
|
|
"mode": "0700"
|
|
},
|
|
{
|
|
"id": "server",
|
|
"type": "container",
|
|
"name": "ollama",
|
|
"image": "ollama/ollama@sha256:32931b46719f673c05fdbaa81ccb26da18ea4a1c57590a754874ab28ba269eb2",
|
|
"network": "host",
|
|
"env": {
|
|
"OLLAMA_HOST": "0.0.0.0:11434"
|
|
},
|
|
"volumes": [
|
|
"/services/ollama:/root/.ollama"
|
|
]
|
|
}
|
|
]
|
|
}
|