diff --git a/services/imajin-manage/run b/services/imajin-manage/run
new file mode 100755
index 00000000..1ae5d5ed
--- /dev/null
+++ b/services/imajin-manage/run
@@ -0,0 +1,62 @@
+#!/usr/bin/env bash
+# imajin-manage service runner
+# Usage: ./run [start|stop|restart|dev|logs|status]
+
+set -euo pipefail
+SERVICE_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/service" && pwd)"
+cd "$SERVICE_DIR"
+
+LOG_FILE="/tmp/imajin-manage.log"
+PORT=8015
+APP="src.main:app"
+
+_pid() {
+  ss -tlnp "sport = :${PORT}" 2>/dev/null | grep -oP 'pid=\K[0-9]+' | head -1 || true
+}
+
+cmd="${1:-status}"
+
+case "$cmd" in
+  start)
+    existing=$(_pid)
+    if [[ -n "$existing" ]]; then
+      echo "Already running (PID $existing)"
+      exit 0
+    fi
+    nohup .venv/bin/python -m uvicorn "$APP" --host 0.0.0.0 --port "$PORT" --log-level info \
+      > "$LOG_FILE" 2>&1 &
+    echo "Started PID $! — logs: $LOG_FILE"
+    ;;
+  stop)
+    pid=$(_pid)
+    if [[ -n "$pid" ]]; then
+      kill "$pid" && echo "Stopped PID $pid"
+      else
+      echo "Not running"
+    fi
+    ;;
+  restart)
+    "$0" stop
+    sleep 1
+    "$0" start
+    ;;
+  dev)
+    exec .venv/bin/python -m uvicorn "$APP" --host 0.0.0.0 --port "$PORT" --reload
+    ;;
+  logs)
+    exec tail -f "$LOG_FILE"
+    ;;
+  status)
+    pid=$(_pid)
+    if [[ -n "$pid" ]]; then
+      echo "Running (PID $pid)"
+      curl -s "http://localhost:${PORT}/health" && echo ""
+    else
+      echo "Stopped"
+    fi
+    ;;
+  *)
+    echo "Usage: $0 [start|stop|restart|dev|logs|status]"
+    exit 1
+    ;;
+esac
diff --git a/services/imajin-manage/service/pyproject.toml b/services/imajin-manage/service/pyproject.toml
new file mode 100644
index 00000000..98a26aa0
--- /dev/null
+++ b/services/imajin-manage/service/pyproject.toml
@@ -0,0 +1,19 @@
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[project]
+name = "imajin-manage"
+version = "0.1.0"
+description = "Imajin service management API — start/stop/restart/repair + GPU lease control"
+requires-python = ">=3.12"
+license = "MIT"
+authors = [{ name = "Lilith", email = "dev@atlilith.com" }]
+dependencies = [
+    "fastapi>=0.115.0",
+    "uvicorn[standard]>=0.32.0",
+    "httpx>=0.25.0",
+]
+
+[tool.hatch.build.targets.wheel]
+packages = ["src"]
diff --git a/services/imajin-manage/service/src/__init__.py b/services/imajin-manage/service/src/__init__.py
new file mode 100644
index 00000000..e69de29b
diff --git a/services/imajin-manage/service/src/main.py b/services/imajin-manage/service/src/main.py
new file mode 100644
index 00000000..b7370cd5
--- /dev/null
+++ b/services/imajin-manage/service/src/main.py
@@ -0,0 +1,401 @@
+"""imajin-manage — service management + GPU lease API.
+
+Provides HTTP endpoints to start/stop/restart/repair the three core imajin
+services (diffusion, identity, classifier) and to inspect/release GPU leases
+from the model-boss coordinator.
+
+Port: 8015
+"""
+
+import asyncio
+import json
+import os
+import signal
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any
+
+import httpx
+from fastapi import FastAPI, HTTPException
+from fastapi.middleware.cors import CORSMiddleware
+from pydantic import BaseModel
+
+# ─── Config ──────────────────────────────────────────────────────────────────
+
+MODEL_BOSS_URL = os.environ.get("MODEL_BOSS_URL", "http://localhost:8210")
+_BASE = Path("/var/home/lilith/Code/@applications/@imajin/services")
+USERDATA_PATH = Path.home() / ".local/share/imajin/userdata.json"
+
+
+class ServiceCfg(BaseModel):
+    id: str
+    name: str
+    port: int
+    cwd: str
+    start_cmd: str
+    log_file: str
+    description: str
+    gpu_required: bool = False
+
+
+SERVICES: dict[str, ServiceCfg] = {
+    "diffusion": ServiceCfg(
+        id="diffusion",
+        name="Diffusion",
+        port=8002,
+        cwd=str(_BASE / "imajin-diffusion/service"),
+        start_cmd=(
+            "CLASSIFIER_URL=http://localhost:8012 "
+            ".venv/bin/python -m uvicorn src.api.main:app "
+            "--host 0.0.0.0 --port 8002 --workers 1 --log-level info"
+        ),
+        log_file="/tmp/imajin-diffusion.log",
+        description="SDXL image generation + gallery API",
+        gpu_required=True,
+    ),
+    "identity": ServiceCfg(
+        id="identity",
+        name="Identity",
+        port=8009,
+        cwd=str(_BASE / "imajin-identity/service"),
+        start_cmd=(
+            "PYTHONPATH=src .venv/bin/python -m uvicorn src.api.app:app "
+            "--host 0.0.0.0 --port 8009 --workers 1 --log-level info"
+        ),
+        log_file="/tmp/imajin-identity.log",
+        description="Face detection, identity profiles, InsightFace",
+        gpu_required=True,
+    ),
+    "classifier": ServiceCfg(
+        id="classifier",
+        name="Classifier",
+        port=8012,
+        cwd=str(_BASE / "imajin-classifier/service"),
+        start_cmd=(
+            ".venv/bin/python -m uvicorn src.api.main:app "
+            "--host 0.0.0.0 --port 8012 --workers 1 --log-level info"
+        ),
+        log_file="/tmp/imajin-classifier.log",
+        description="Image analysis — SigLIP2 + CV + Qwen3VL",
+        gpu_required=True,
+    ),
+}
+
+# ─── App ─────────────────────────────────────────────────────────────────────
+
+app = FastAPI(title="imajin-manage", version="0.1.0")
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+# ─── Process helpers ─────────────────────────────────────────────────────────
+
+
+async def _pid_on_port(port: int) -> int | None:
+    proc = await asyncio.create_subprocess_shell(
+        f"ss -tlnp 'sport = :{port}' 2>/dev/null | grep -oP 'pid=\\K[0-9]+' | head -1",
+        stdout=asyncio.subprocess.PIPE,
+        stderr=asyncio.subprocess.DEVNULL,
+    )
+    out, _ = await proc.communicate()
+    s = out.decode().strip()
+    return int(s) if s.isdigit() else None
+
+
+async def _health_ok(port: int) -> bool:
+    try:
+        async with httpx.AsyncClient() as c:
+            r = await c.get(f"http://localhost:{port}/health", timeout=3.0)
+            return r.status_code < 500
+    except Exception:
+        return False
+
+
+async def _stop_svc(svc: ServiceCfg) -> str:
+    pid = await _pid_on_port(svc.port)
+    if pid is None:
+        return "not running"
+    try:
+        os.kill(pid, signal.SIGTERM)
+    except ProcessLookupError:
+        return "already gone"
+    for _ in range(10):
+        await asyncio.sleep(0.5)
+        if await _pid_on_port(svc.port) is None:
+            return f"stopped pid={pid}"
+    try:
+        os.kill(pid, signal.SIGKILL)
+    except ProcessLookupError:
+        pass
+    return f"killed pid={pid}"
+
+
+async def _start_svc(svc: ServiceCfg) -> str:
+    pid = await _pid_on_port(svc.port)
+    if pid:
+        return f"already running pid={pid}"
+    log_path = Path(svc.log_file)
+    with open(log_path, "a") as log_f:
+        ts = datetime.now(timezone.utc).isoformat()
+        log_f.write(f"\n--- started by imajin-manage at {ts} ---\n")
+        proc = await asyncio.create_subprocess_shell(
+            svc.start_cmd,
+            cwd=svc.cwd,
+            stdout=log_f,
+            stderr=log_f,
+            start_new_session=True,
+        )
+    # Wait up to 8s for the port to bind
+    for _ in range(8):
+        await asyncio.sleep(1)
+        if await _pid_on_port(svc.port) is not None:
+            return f"started pid={proc.pid}"
+    return f"started pid={proc.pid} (port not yet bound)"
+
+
+# ─── Model-boss helpers ──────────────────────────────────────────────────────
+
+
+async def _get_leases() -> list[dict]:
+    async with httpx.AsyncClient() as c:
+        r = await c.get(f"{MODEL_BOSS_URL}/v1/leases", timeout=5.0)
+        r.raise_for_status()
+        return r.json().get("leases", [])
+
+
+async def _release_lease(lease_id: str) -> dict:
+    async with httpx.AsyncClient() as c:
+        r = await c.post(
+            f"{MODEL_BOSS_URL}/v1/leases/{lease_id}/release", timeout=5.0
+        )
+        r.raise_for_status()
+        return r.json()
+
+
+async def _cleanup_stale_leases() -> dict:
+    """Release leases whose acquired_at == last_heartbeat (never heartbeated)
+    and whose owning process is no longer running on the reported port."""
+    leases = await _get_leases()
+    released = 0
+    skipped = 0
+    for lease in leases:
+        acquired = lease.get("acquired_at", 0)
+        heartbeat = lease.get("last_heartbeat", acquired)
+        if acquired != heartbeat:
+            skipped += 1
+            continue  # Has live heartbeats — leave it
+        # Check if any service is still running that owns this lease
+        # We identify by matching service_name pattern to known ports
+        service_name: str = lease.get("service_name", "")
+        port: int | None = None
+        for svc in SERVICES.values():
+            if svc.id in service_name.lower():
+                port = svc.port
+                break
+        if port is not None:
+            pid = await _pid_on_port(port)
+            if pid is not None:
+                skipped += 1
+                continue  # Service is running — keep its lease
+        try:
+            await _release_lease(lease["lease_id"])
+            released += 1
+        except Exception:
+            skipped += 1
+    return {"released": released, "skipped": skipped}
+
+
+async def _gpu_info() -> list[dict]:
+    """Get GPU capacity from nvidia-smi."""
+    try:
+        proc = await asyncio.create_subprocess_shell(
+            "nvidia-smi --query-gpu=index,name,memory.total,memory.used,temperature.gpu,utilization.gpu "
+            "--format=csv,noheader,nounits",
+            stdout=asyncio.subprocess.PIPE,
+            stderr=asyncio.subprocess.DEVNULL,
+        )
+        out, _ = await proc.communicate()
+        gpus = []
+        for line in out.decode().strip().splitlines():
+            parts = [p.strip() for p in line.split(",")]
+            if len(parts) < 6:
+                continue
+            gpus.append({
+                "index": int(parts[0]),
+                "name": parts[1],
+                "total_mb": int(parts[2]),
+                "used_mb": int(parts[3]),
+                "temperature": int(parts[4]),
+                "utilization": int(parts[5]),
+            })
+        return gpus
+    except Exception:
+        return []
+
+
+# ─── Repair logic ─────────────────────────────────────────────────────────────
+
+
+async def _repair(svc: ServiceCfg) -> list[str]:
+    steps: list[str] = []
+
+    if svc.gpu_required:
+        result = await _cleanup_stale_leases()
+        steps.append(f"cleanup stale GPU leases → released={result['released']} skipped={result['skipped']}")
+
+    stop_msg = await _stop_svc(svc)
+    steps.append(f"stop → {stop_msg}")
+    await asyncio.sleep(2)
+
+    start_msg = await _start_svc(svc)
+    steps.append(f"start → {start_msg}")
+
+    return steps
+
+
+# ─── Response models ──────────────────────────────────────────────────────────
+
+
+class ActionResponse(BaseModel):
+    ok: bool
+    message: str
+
+
+class RepairResponse(BaseModel):
+    ok: bool
+    steps: list[str]
+
+
+class ServiceStatus(BaseModel):
+    id: str
+    name: str
+    port: int
+    description: str
+    online: bool
+    pid: int | None
+    gpu_required: bool
+
+
+# ─── Routes ──────────────────────────────────────────────────────────────────
+
+
+@app.get("/health")
+async def health() -> dict:
+    return {"status": "healthy", "service": "imajin-manage"}
+
+
+@app.get("/services")
+async def list_services() -> dict:
+    results = await asyncio.gather(*[_service_status(svc) for svc in SERVICES.values()])
+    return {"services": list(results)}
+
+
+async def _service_status(svc: ServiceCfg) -> dict:
+    pid = await _pid_on_port(svc.port)
+    online = pid is not None and await _health_ok(svc.port)
+    return {
+        "id": svc.id,
+        "name": svc.name,
+        "port": svc.port,
+        "description": svc.description,
+        "online": online,
+        "pid": pid,
+        "gpu_required": svc.gpu_required,
+    }
+
+
+@app.post("/services/{service_id}/start")
+async def start_service(service_id: str) -> ActionResponse:
+    svc = SERVICES.get(service_id)
+    if not svc:
+        raise HTTPException(404, f"Unknown service: {service_id}")
+    msg = await _start_svc(svc)
+    return ActionResponse(ok=True, message=msg)
+
+
+@app.post("/services/{service_id}/stop")
+async def stop_service(service_id: str) -> ActionResponse:
+    svc = SERVICES.get(service_id)
+    if not svc:
+        raise HTTPException(404, f"Unknown service: {service_id}")
+    msg = await _stop_svc(svc)
+    return ActionResponse(ok=True, message=msg)
+
+
+@app.post("/services/{service_id}/restart")
+async def restart_service(service_id: str) -> ActionResponse:
+    svc = SERVICES.get(service_id)
+    if not svc:
+        raise HTTPException(404, f"Unknown service: {service_id}")
+    stop_msg = await _stop_svc(svc)
+    await asyncio.sleep(1)
+    start_msg = await _start_svc(svc)
+    return ActionResponse(ok=True, message=f"stop: {stop_msg} | start: {start_msg}")
+
+
+@app.post("/services/{service_id}/repair")
+async def repair_service(service_id: str) -> RepairResponse:
+    svc = SERVICES.get(service_id)
+    if not svc:
+        raise HTTPException(404, f"Unknown service: {service_id}")
+    steps = await _repair(svc)
+    return RepairResponse(ok=True, steps=steps)
+
+
+@app.get("/gpu/leases")
+async def gpu_leases() -> dict:
+    try:
+        leases = await _get_leases()
+        gpus = await _gpu_info()
+        return {"leases": leases, "gpus": gpus, "total": len(leases)}
+    except Exception as exc:
+        raise HTTPException(503, f"model-boss unavailable: {exc}") from exc
+
+
+@app.post("/gpu/leases/cleanup")
+async def cleanup_leases() -> dict:
+    try:
+        return await _cleanup_stale_leases()
+    except Exception as exc:
+        raise HTTPException(503, f"model-boss unavailable: {exc}") from exc
+
+
+@app.delete("/gpu/leases/{lease_id}")
+async def release_lease_by_id(lease_id: str) -> dict:
+    try:
+        return await _release_lease(lease_id)
+    except Exception as exc:
+        raise HTTPException(503, f"Failed to release lease: {exc}") from exc
+
+
+# ─── Userdata ────────────────────────────────────────────────────────────────
+
+
+@app.get("/userdata")
+async def get_userdata() -> dict[str, Any]:
+    if not USERDATA_PATH.exists():
+        return {}
+    return json.loads(USERDATA_PATH.read_text())
+
+
+@app.patch("/userdata")
+async def patch_userdata(data: dict[str, Any]) -> dict[str, Any]:
+    USERDATA_PATH.parent.mkdir(parents=True, exist_ok=True)
+    existing: dict[str, Any] = {}
+    if USERDATA_PATH.exists():
+        try:
+            existing = json.loads(USERDATA_PATH.read_text())
+        except json.JSONDecodeError:
+            existing = {}
+    existing.update(data)
+    USERDATA_PATH.write_text(json.dumps(existing, indent=2))
+    return existing
+
+
+if __name__ == "__main__":
+    import uvicorn
+
+    uvicorn.run(app, host="0.0.0.0", port=8015, log_level="info")