diff --git a/autogpt_platform/backend/backend/api/rest_api.py b/autogpt_platform/backend/backend/api/rest_api.py
index 3bd50c78ba..6984b1b28f 100644
--- a/autogpt_platform/backend/backend/api/rest_api.py
+++ b/autogpt_platform/backend/backend/api/rest_api.py
@@ -394,6 +394,11 @@ app.include_router(
     tags=["v2", "llm"],
     prefix="/api",
 )
+app.include_router(
+    backend.server.v2.llm.admin_router,
+    tags=["v2", "llm", "admin"],
+    prefix="/api",
+)
 
 app.mount("/external-api", external_api)
 
diff --git a/autogpt_platform/backend/backend/server/v2/llm/__init__.py b/autogpt_platform/backend/backend/server/v2/llm/__init__.py
index c69da22392..cf3d75f7ba 100644
--- a/autogpt_platform/backend/backend/server/v2/llm/__init__.py
+++ b/autogpt_platform/backend/backend/server/v2/llm/__init__.py
@@ -1,5 +1,6 @@
-"""LLM registry public API."""
+"""LLM registry API (public + admin)."""
 
+from .admin_routes import router as admin_router
 from .routes import router
 
-__all__ = ["router"]
+__all__ = ["router", "admin_router"]
diff --git a/autogpt_platform/backend/backend/server/v2/llm/admin_model.py b/autogpt_platform/backend/backend/server/v2/llm/admin_model.py
new file mode 100644
index 0000000000..fab77c2520
--- /dev/null
+++ b/autogpt_platform/backend/backend/server/v2/llm/admin_model.py
@@ -0,0 +1,84 @@
+"""Request/response models for LLM registry admin API."""
+
+from typing import Any
+
+from pydantic import BaseModel, Field
+
+
+class CreateLlmProviderRequest(BaseModel):
+    """Request model for creating an LLM provider."""
+
+    name: str = Field(..., description="Provider identifier (e.g., 'openai', 'anthropic')")
+    display_name: str = Field(..., description="Human-readable provider name")
+    description: str | None = Field(None, description="Provider description")
+    default_credential_provider: str | None = Field(
+        None, description="Default credential system identifier"
+    )
+    default_credential_id: str | None = Field(None, description="Default credential ID")
+    default_credential_type: str | None = Field(None, description="Default credential type")
+    metadata: dict[str, Any] = Field(default_factory=dict, description="Additional metadata")
+
+
+class UpdateLlmProviderRequest(BaseModel):
+    """Request model for updating an LLM provider."""
+
+    display_name: str | None = Field(None, description="Human-readable provider name")
+    description: str | None = Field(None, description="Provider description")
+    default_credential_provider: str | None = Field(
+        None, description="Default credential system identifier"
+    )
+    default_credential_id: str | None = Field(None, description="Default credential ID")
+    default_credential_type: str | None = Field(None, description="Default credential type")
+    metadata: dict[str, Any] | None = Field(None, description="Additional metadata")
+
+
+class CreateLlmModelRequest(BaseModel):
+    """Request model for creating an LLM model."""
+
+    slug: str = Field(..., description="Model slug (e.g., 'gpt-4', 'claude-3-opus')")
+    display_name: str = Field(..., description="Human-readable model name")
+    description: str | None = Field(None, description="Model description")
+    provider_id: str = Field(..., description="Provider ID (UUID)")
+    creator_id: str | None = Field(None, description="Creator ID (UUID)")
+    context_window: int = Field(..., description="Maximum context window in tokens", gt=0)
+    max_output_tokens: int | None = Field(
+        None, description="Maximum output tokens (None if unlimited)", gt=0
+    )
+    price_tier: int = Field(..., description="Price tier (1=cheapest, 2=medium, 3=expensive)", ge=1, le=3)
+    is_enabled: bool = Field(default=True, description="Whether the model is enabled")
+    is_recommended: bool = Field(default=False, description="Whether the model is recommended")
+    supports_tools: bool = Field(default=False, description="Supports function calling")
+    supports_json_output: bool = Field(default=False, description="Supports JSON output mode")
+    supports_reasoning: bool = Field(default=False, description="Supports reasoning mode")
+    supports_parallel_tool_calls: bool = Field(default=False, description="Supports parallel tool calls")
+    capabilities: dict[str, Any] = Field(
+        default_factory=dict, description="Additional capabilities"
+    )
+    metadata: dict[str, Any] = Field(default_factory=dict, description="Additional metadata")
+
+
+class UpdateLlmModelRequest(BaseModel):
+    """Request model for updating an LLM model."""
+
+    display_name: str | None = Field(None, description="Human-readable model name")
+    description: str | None = Field(None, description="Model description")
+    creator_id: str | None = Field(None, description="Creator ID (UUID)")
+    context_window: int | None = Field(
+        None, description="Maximum context window in tokens", gt=0
+    )
+    max_output_tokens: int | None = Field(
+        None, description="Maximum output tokens (None if unlimited)", gt=0
+    )
+    price_tier: int | None = Field(
+        None, description="Price tier (1=cheapest, 2=medium, 3=expensive)", ge=1, le=3
+    )
+    is_enabled: bool | None = Field(None, description="Whether the model is enabled")
+    is_recommended: bool | None = Field(None, description="Whether the model is recommended")
+    supports_tools: bool | None = Field(None, description="Supports function calling")
+    supports_json_output: bool | None = Field(None, description="Supports JSON output mode")
+    supports_reasoning: bool | None = Field(None, description="Supports reasoning mode")
+    supports_parallel_tool_calls: bool | None = Field(
+        None, description="Supports parallel tool calls"
+    )
+    capabilities: dict[str, Any] | None = Field(None, description="Additional capabilities")
+    metadata: dict[str, Any] | None = Field(None, description="Additional metadata")
diff --git a/autogpt_platform/backend/backend/server/v2/llm/admin_routes.py b/autogpt_platform/backend/backend/server/v2/llm/admin_routes.py
new file mode 100644
index 0000000000..29aee4c86e
--- /dev/null
+++ b/autogpt_platform/backend/backend/server/v2/llm/admin_routes.py
@@ -0,0 +1,121 @@
+"""Admin write API for LLM registry management.
+
+Provides endpoints for creating, updating, and deleting:
+- Models
+- Providers
+- Costs
+- Creators
+- Migrations
+
+All endpoints require admin authentication.
+"""
+
+from typing import Any
+
+import autogpt_libs.auth
+from fastapi import APIRouter, HTTPException, Security, status
+
+from backend.server.v2.llm.admin_model import (
+    CreateLlmModelRequest,
+    CreateLlmProviderRequest,
+    UpdateLlmModelRequest,
+    UpdateLlmProviderRequest,
+)
+
+router = APIRouter()
+
+
+@router.post(
+    "/llm/models",
+    status_code=status.HTTP_201_CREATED,
+    dependencies=[Security(autogpt_libs.auth.requires_admin_user)],
+)
+async def create_model(
+    request: CreateLlmModelRequest,
+) -> dict[str, Any]:
+    """Create a new LLM model.
+
+    Requires admin authentication.
+    """
+    # TODO: Implement model creation
+    raise HTTPException(
+        status_code=status.HTTP_501_NOT_IMPLEMENTED,
+        detail="Model creation not yet implemented",
+    )
+
+
+@router.patch("/llm/models/{slug}", dependencies=[Security(autogpt_libs.auth.requires_admin_user)])
+async def update_model(
+    slug: str,
+    request: UpdateLlmModelRequest,
+) -> dict[str, Any]:
+    """Update an existing LLM model.
+
+    Requires admin authentication.
+    """
+    # TODO: Implement model update
+    raise HTTPException(
+        status_code=status.HTTP_501_NOT_IMPLEMENTED,
+        detail="Model update not yet implemented",
+    )
+
+
+@router.delete("/llm/models/{slug}", dependencies=[Security(autogpt_libs.auth.requires_admin_user)], status_code=status.HTTP_204_NO_CONTENT)
+async def delete_model(
+    slug: str,
+) -> None:
+    """Delete an LLM model.
+
+    Requires admin authentication.
+    """
+    # TODO: Implement model deletion
+    raise HTTPException(
+        status_code=status.HTTP_501_NOT_IMPLEMENTED,
+        detail="Model deletion not yet implemented",
+    )
+
+
+@router.post("/llm/providers", status_code=status.HTTP_201_CREATED)
+async def create_provider(
+    request: CreateLlmProviderRequest,
+) -> dict[str, Any]:
+    """Create a new LLM provider.
+
+    Requires admin authentication.
+    """
+    # TODO: Implement provider creation
+    raise HTTPException(
+        status_code=status.HTTP_501_NOT_IMPLEMENTED,
+        detail="Provider creation not yet implemented",
+    )
+
+
+@router.patch("/llm/providers/{name}", dependencies=[Security(autogpt_libs.auth.requires_admin_user)])
+async def update_provider(
+    name: str,
+    request: UpdateLlmProviderRequest,
+) -> dict[str, Any]:
+    """Update an existing LLM provider.
+
+    Requires admin authentication.
+    """
+    # TODO: Implement provider update
+    raise HTTPException(
+        status_code=status.HTTP_501_NOT_IMPLEMENTED,
+        detail="Provider update not yet implemented",
+    )
+
+
+@router.delete("/llm/providers/{name}", dependencies=[Security(autogpt_libs.auth.requires_admin_user)], status_code=status.HTTP_204_NO_CONTENT)
+async def delete_provider(
+    name: str,
+) -> None:
+    """Delete an LLM provider.
+
+    Requires admin authentication.
+    """
+    # TODO: Implement provider deletion
+    raise HTTPException(
+        status_code=status.HTTP_501_NOT_IMPLEMENTED,
+        detail="Provider deletion not yet implemented",
+    )