Skip to content

dashboard: rename Ollama endpoints to /api/llm, remove dead registry browse #67

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
AlienWalker1995 merged 1 commit into from
Jul 1, 2026
Merged

dashboard: rename Ollama endpoints to /api/llm, remove dead registry browse #67

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 4 additions & 4 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -131,16 +131,16 @@ Auto-generated: **`overrides/compute.yml`** (from hardware detection). Do not co

### Dashboard

Reach the dashboard at `https://${CADDY_TAILNET_HOSTNAME}/dash/` (Google SSO front door; allowlist via `auth/oauth2-proxy/emails.txt`). It lists models (Ollama and ComfyUI), links to other services, dependency health, and searchable model pulls. **`OPS_CONTROLLER_TOKEN`** lets it restart services and run **`POST /api/comfyui/install-node-requirements`**. **`DASHBOARD_AUTH_TOKEN`** is an optional bearer layer for non-browser API access; the browser path is gated by SSO at the proxy level.
Reach the dashboard at `https://${CADDY_TAILNET_HOSTNAME}/dash/` (Google SSO front door; allowlist via `auth/oauth2-proxy/emails.txt`). It lists models (GGUF/llama.cpp and ComfyUI), links to other services, dependency health, and Hugging Face model pulls. **`OPS_CONTROLLER_TOKEN`** lets it restart services and run **`POST /api/comfyui/install-node-requirements`**. **`DASHBOARD_AUTH_TOKEN`** is an optional bearer layer for non-browser API access; the browser path is gated by SSO at the proxy level.

After code changes affecting the dashboard image: `.\compose.ps1 build dashboard` then `.\compose.ps1 up -d` (or `./compose` equivalents).

### Ollama models
### LLM models (GGUF / llama.cpp)

Pull lists and defaults come from **`.env`** (`MODELS`, `DEFAULT_MODEL`). Pull via the dashboard or:
The stack pulls GGUF files (served by llama.cpp) directly from Hugging Face. Repo lists and defaults come from **`.env`** (`GGUF_MODELS`, `DEFAULT_MODEL`). Pull via the dashboard's **Models** panel (enter a Hugging Face repo id, a `huggingface.co/…`/`.gguf` URL, or `.env` to pull all `GGUF_MODELS`), or from the CLI:

```bash
./compose run --rm model-puller
./compose run --rm gguf-puller
```

### ComfyUI (LTX-2)
Expand Down
202 changes: 63 additions & 139 deletions dashboard/app.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -210,93 +210,17 @@ def _model_gateway_headers() -> dict[str, str]:
headers["Authorization"] = f"Bearer {MODEL_GATEWAY_API_KEY}"
return headers

# Ollama library: fetched from community JSON (all pullable model:tag names)
OLLAMA_LIBRARY_URL = os.environ.get(
"OLLAMA_LIBRARY_URL",
"https://yuma-shintani.github.io/ollama-model-library/model.json",
)
OLLAMA_LIBRARY_CACHE_TTL = float(os.environ.get("OLLAMA_LIBRARY_CACHE_TTL_SEC", "86400")) # 24h
_ollama_library_cache: list[str] = []
_ollama_library_ts: float = 0.0

# Fallback when fetch fails (minimal curated list)
OLLAMA_LIBRARY_FALLBACK = [
"llama3.2", "llama3.1", "deepseek-r1:7b", "qwen2.5:7b", "qwen3:14b", "qwen3:14b-q4_K_M",
"mistral", "nomic-embed-text", "phi4", "gemma3",
]

# Background pull status dicts
_comfyui_status: dict = {"running": False, "output": "", "done": False, "success": None}
_ollama_pull_status: dict = {"running": False, "model": "", "output": "", "pct": 0, "done": False, "success": None}
_gguf_pull_status: dict = {"running": False, "model": "", "output": "", "pct": 0, "done": False, "success": None}



class PullRequest(BaseModel):
model: str


# --- Ollama ---


def _fetch_ollama_library() -> list[str]:
"""Fetch pullable model names from Ollama registry. Uses community JSON; caches 24h."""
global _ollama_library_cache, _ollama_library_ts
now = time.monotonic()
with _state_lock:
if _ollama_library_cache and (now - _ollama_library_ts) < OLLAMA_LIBRARY_CACHE_TTL:
return list(_ollama_library_cache)

urls = [OLLAMA_LIBRARY_URL]
for url in urls:
try:
req = urllib.request.Request(url, headers={"Accept": "application/json"})
with urllib.request.urlopen(req, timeout=15) as resp:
data = json.loads(resp.read().decode())
except Exception as e:
logger.warning("Ollama library fetch failed from %s: %s", url, e)
continue

names: set[str] = set()
if isinstance(data, list):
# yuma-shintani format: [{"name":"llama3.1","tags":[{"name":"llama3.1:8b"},...]}, ...]
for item in data:
if isinstance(item, dict):
base = (item.get("name") or "").strip()
tags = item.get("tags") or []
for t in tags:
if isinstance(t, dict) and t.get("name"):
names.add(str(t["name"]).strip())
if base:
names.add(base) # e.g. llama3.1 -> llama3.1:latest
elif isinstance(data, dict):
# Official format: {"library": {"llama3.1": {"tags": ["8b","70b"]}, ...}}
lib = data.get("library") or data
if isinstance(lib, dict):
for base, meta in lib.items():
if isinstance(meta, dict):
for tag in meta.get("tags") or []:
names.add(f"{base}:{tag}" if tag else base)
else:
names.add(base)

if names:
result = sorted(names)
with _state_lock:
_ollama_library_cache = result
_ollama_library_ts = now
return result

with _state_lock:
_ollama_library_cache = OLLAMA_LIBRARY_FALLBACK
_ollama_library_ts = now
return list(OLLAMA_LIBRARY_FALLBACK)


@app.get("/api/ollama/library")
async def ollama_library():
"""List models available in the Ollama registry (fetched programmatically, cached 24h)."""
models = await asyncio.to_thread(_fetch_ollama_library)
return {"models": models, "ok": True}
# --- LLM (llama.cpp / GGUF) ---


_GGUF_MODELS_DIR = Path(os.environ.get("GGUF_MODELS_DIR", "/gguf-models"))
Expand All @@ -315,8 +239,8 @@ def _scan_gguf_models() -> list[dict]:
return models


@app.get("/api/ollama/models")
async def ollama_models():
@app.get("/api/llm/models")
async def llm_models():
"""List GGUF models available on disk (primary) merged with gateway active-model info."""
disk_models = await asyncio.to_thread(_scan_gguf_models)
if disk_models:
Expand All @@ -332,8 +256,8 @@ async def ollama_models():
return {"models": [], "ok": False, "error": str(e)}


@app.post("/api/ollama/delete")
async def ollama_delete(req: PullRequest):
@app.post("/api/llm/delete")
async def llm_delete(req: PullRequest):
"""Delete a GGUF model file from disk."""
name = (req.model or "").strip()
if not name or ".." in name or "/" in name:
Expand All @@ -355,8 +279,8 @@ async def ollama_delete(req: PullRequest):
return {"ok": True, "message": f"Deleted '{name}' from disk."}


@app.post("/api/ollama/unload")
async def ollama_unload(req: PullRequest):
@app.post("/api/llm/unload")
async def llm_unload(req: PullRequest):
"""Unload the currently active model from the gateway without deleting GGUF files."""
name = (req.model or "").strip()
if not name or ".." in name:
Expand All @@ -376,7 +300,7 @@ async def ollama_unload(req: PullRequest):
except HTTPException:
raise
except Exception as e:
raise HTTPException(status_code=502, detail=f"Ollama request failed: {e}") from e
raise HTTPException(status_code=502, detail=f"Model gateway request failed: {e}") from e


@app.post("/api/llamacpp/switch")
Expand Down Expand Up @@ -452,38 +376,38 @@ async def _do_set_active_model(req: PullRequest, request: Request):
return {"ok": all_ok, "model": model, "errors": errors, **results}


def _run_ollama_pull(model: str):
def _run_gguf_pull(model: str):
"""Download GGUFs via ops-controller gguf-puller (docker compose --profile models)."""
global _ollama_pull_status
global _gguf_pull_status
with _state_lock:
_ollama_pull_status = {"running": True, "model": model, "output": "", "pct": 0, "done": False, "success": None}
_gguf_pull_status = {"running": True, "model": model, "output": "", "pct": 0, "done": False, "success": None}

repos = _normalize_gguf_pull_repos(model)
if repos is None:
repos = _normalize_gguf_pull_repos(_hf_url_to_ollama(model))
repos = _normalize_gguf_pull_repos(_hf_url_to_repo(model))
if repos is None:
msg = (
"This stack uses GGUF files (llama.cpp), not the Ollama registry.\n\n"
"This stack pulls GGUF files (llama.cpp) directly from Hugging Face.\n\n"
"Enter a Hugging Face repo id (e.g. bartowski/Llama-3.2-3B-Instruct-GGUF), "
"a huggingface.co/… page or .gguf URL, hf.co/owner/repo, or type .env to pull all "
"repos listed in GGUF_MODELS in your .env.\n\n"
"Names like llama3.2:8b only work with a real Ollama daemon, not this gateway."
"Bare tag names like llama3.2:8b are not supported; use a Hugging Face repo id or .gguf URL."
)
with _state_lock:
_ollama_pull_status["output"] = msg
_ollama_pull_status["success"] = False
_ollama_pull_status["running"] = False
_ollama_pull_status["done"] = True
_gguf_pull_status["output"] = msg
_gguf_pull_status["success"] = False
_gguf_pull_status["running"] = False
_gguf_pull_status["done"] = True
return

ops_url = os.environ.get("OPS_CONTROLLER_URL", "http://ops-controller:9000").rstrip("/")
token = os.environ.get("OPS_CONTROLLER_TOKEN", "").strip()
if not token:
with _state_lock:
_ollama_pull_status["output"] = "OPS_CONTROLLER_TOKEN is not set; cannot run gguf-puller from the dashboard."
_ollama_pull_status["success"] = False
_ollama_pull_status["running"] = False
_ollama_pull_status["done"] = True
_gguf_pull_status["output"] = "OPS_CONTROLLER_TOKEN is not set; cannot run gguf-puller from the dashboard."
_gguf_pull_status["success"] = False
_gguf_pull_status["running"] = False
_gguf_pull_status["done"] = True
return

try:
Expand All @@ -496,21 +420,21 @@ def _run_ollama_pull(model: str):
)
if r.status_code == 409:
with _state_lock:
_ollama_pull_status["output"] = "Another model or GGUF pull is already in progress."
_ollama_pull_status["success"] = False
_ollama_pull_status["running"] = False
_ollama_pull_status["done"] = True
_gguf_pull_status["output"] = "Another model or GGUF pull is already in progress."
_gguf_pull_status["success"] = False
_gguf_pull_status["running"] = False
_gguf_pull_status["done"] = True
return
if r.status_code >= 400:
try:
det = r.json().get("detail", r.text)
except (ValueError, UnicodeDecodeError):
det = r.text
with _state_lock:
_ollama_pull_status["output"] = f"Failed to start gguf-puller: {det}"
_ollama_pull_status["success"] = False
_ollama_pull_status["running"] = False
_ollama_pull_status["done"] = True
_gguf_pull_status["output"] = f"Failed to start gguf-puller: {det}"
_gguf_pull_status["success"] = False
_gguf_pull_status["running"] = False
_gguf_pull_status["done"] = True
return

deadline = time.time() + 7200 # 2-hour max
Expand All @@ -536,44 +460,44 @@ def _run_ollama_pull(model: str):
raise RuntimeError(f"Poll failed 20 times: {poll_err}")
continue
with _state_lock:
_ollama_pull_status["output"] = st.get("output", "")
_ollama_pull_status["pct"] = 50 if st.get("running") else 100
_gguf_pull_status["output"] = st.get("output", "")
_gguf_pull_status["pct"] = 50 if st.get("running") else 100
if st.get("done"):
with _state_lock:
_ollama_pull_status["success"] = bool(st.get("success"))
_ollama_pull_status["running"] = False
_ollama_pull_status["done"] = True
_gguf_pull_status["success"] = bool(st.get("success"))
_gguf_pull_status["running"] = False
_gguf_pull_status["done"] = True
break
else:
raise TimeoutError("GGUF pull timed out after 2 hours")
except Exception as e:
logger.error("GGUF pull failed: %s", e)
with _state_lock:
_ollama_pull_status["output"] = (_ollama_pull_status.get("output") or "") + f"\nError: {e}"
_ollama_pull_status["success"] = False
_ollama_pull_status["running"] = False
_ollama_pull_status["done"] = True
_gguf_pull_status["output"] = (_gguf_pull_status.get("output") or "") + f"\nError: {e}"
_gguf_pull_status["success"] = False
_gguf_pull_status["running"] = False
_gguf_pull_status["done"] = True


@app.post("/api/ollama/pull")
async def ollama_pull(req: PullRequest):
"""Start GGUF download (gguf-puller via ops-controller) in background. Poll /api/ollama/pull/status."""
global _ollama_pull_status
@app.post("/api/llm/pull")
async def llm_pull(req: PullRequest):
"""Start GGUF download (gguf-puller via ops-controller) in background. Poll /api/llm/pull/status."""
global _gguf_pull_status
with _state_lock:
if _ollama_pull_status.get("running"):
if _gguf_pull_status.get("running"):
raise HTTPException(status_code=409, detail="Pull already in progress")
_ollama_pull_status["running"] = True
_ollama_pull_status["model"] = req.model
thread = threading.Thread(target=_run_ollama_pull, args=(req.model,), daemon=True)
_gguf_pull_status["running"] = True
_gguf_pull_status["model"] = req.model
thread = threading.Thread(target=_run_gguf_pull, args=(req.model,), daemon=True)
thread.start()
return {"status": "started", "model": req.model}


@app.get("/api/ollama/pull/status")
async def ollama_pull_status():
"""Get Ollama pull progress."""
@app.get("/api/llm/pull/status")
async def llm_pull_status():
"""Get GGUF pull progress."""
with _state_lock:
return dict(_ollama_pull_status)
return dict(_gguf_pull_status)


# --- ComfyUI ---
Expand Down Expand Up @@ -957,7 +881,7 @@ class ModelPullRequest(BaseModel):
def _normalize_gguf_pull_repos(model: str) -> str | None:
"""Return comma-separated Hugging Face repo ids for gguf-puller, or '' to use .env GGUF_MODELS.

None means the string is not suitable (e.g. Ollama-style ``llama3.2:8b``).
None means the string is not suitable (e.g. a bare tag like ``llama3.2:8b``).
"""
def _normalize_repo_ref(raw: str) -> str | None:
candidate = raw.strip()
Expand Down Expand Up @@ -999,8 +923,8 @@ def _normalize_repo_ref(raw: str) -> str | None:
return _normalize_repo_ref(s)


def _hf_url_to_ollama(raw: str) -> str:
"""Convert a HuggingFace GGUF URL to Ollama's hf.co/owner/repo format.
def _hf_url_to_repo(raw: str) -> str:
"""Convert a HuggingFace GGUF URL to hf.co/owner/repo form for the gguf-puller.
Non-HF strings (model names, hf.co/ refs) are returned as-is.
"""
if "huggingface.co/" in raw:
Expand All @@ -1016,7 +940,7 @@ def _hf_url_to_ollama(raw: str) -> str:
@app.post("/api/models/download")
async def models_download(req: ModelDownloadRequest, request: Request):
"""Unified model download.
- GGUF / HF repo → background gguf-puller via ops (same as ``/api/ollama/pull``); poll ``/api/ollama/pull/status``.
- GGUF / HF repo → background gguf-puller via ops (same as ``/api/llm/pull``); poll ``/api/llm/pull/status``.
- safetensors / ckpt / pt / bin → proxied to ops-controller for file download.
"""
raw = req.url.strip()
Expand All @@ -1039,15 +963,15 @@ async def models_download(req: ModelDownloadRequest, request: Request):
return {**data, "target": "comfyui"}
else:
with _state_lock:
if _ollama_pull_status.get("running"):
if _gguf_pull_status.get("running"):
raise HTTPException(status_code=409, detail="Pull already in progress")
_ollama_pull_status["running"] = True
thread = threading.Thread(target=_run_ollama_pull, args=(raw,), daemon=True)
_gguf_pull_status["running"] = True
thread = threading.Thread(target=_run_gguf_pull, args=(raw,), daemon=True)
thread.start()
return {
"status": "started",
"target": "gguf",
"message": "Poll /api/ollama/pull/status for progress.",
"message": "Poll /api/llm/pull/status for progress.",
}


Expand Down Expand Up @@ -1571,8 +1495,8 @@ async def performance_summary():
}


@app.get("/api/ollama/ps")
async def ollama_ps():
@app.get("/api/llm/ps")
async def llm_ps():
"""List models currently advertised by model-gateway."""
try:
r = await _get_http_client().get(
Expand Down Expand Up @@ -1814,7 +1738,7 @@ def _open_webui_default_model(name: str) -> str:
@app.post("/api/config/default-model")
async def set_default_model(req: DefaultModelRequest, request: Request):
"""Write DEFAULT_MODEL and OPEN_WEBUI_DEFAULT_MODEL to .env and recreate open-webui."""
# Ollama allows namespaced ids: owner/model:tag (slashes required). Only reject empty / traversal.
# Model ids may be namespaced: owner/model:tag (slashes allowed). Only reject empty / traversal.
name = (req.model or "").strip()
if not name or ".." in name:
raise HTTPException(status_code=400, detail="Invalid model name")
Expand Down
Loading
Loading