Nimblesite vs raw LLM APIs

Short version: Raw LLM APIs are stateless. Every call is a fresh start. Nimblesite is the agent that wraps all of that — memory, loop, tool dispatch, multi-tenancy — already wired up.

What a raw LLM call looks like

Here's what you write when you talk to Claude or GPT directly:

# Every call is a fresh start. You send the full history every time.
history = load_messages_from_db(conversation_id)
history.append({"role": "user", "content": "Hello"})

response = anthropic.messages.create(
    model="claude-sonnet-4-6",
    messages=history,
    tools=my_tools,
)

# If the response is a tool call, parse it, run it,
# append the result to history, and call again.
while response.stop_reason == "tool_use":
    for block in response.content:
        if block.type == "tool_use":
            result = run_tool(block.name, block.input)
            history.append({"role": "assistant", "content": response.content})
            history.append({
                "role": "user",
                "content": [{
                    "type": "tool_result",
                    "tool_use_id": block.id,
                    "content": result,
                }],
            })

    response = anthropic.messages.create(
        model="claude-sonnet-4-6",
        messages=history,
        tools=my_tools,
    )

# Save the final history back to the database.
save_messages_to_db(conversation_id, history)

return response.content[-1].text

Now multiply that by:

• Multi-tenancy (each tenant has their own API key and their own messages table scope)
• Prompt templating (each tenant has their own system prompt)
• Provider failover (what if Anthropic is down?)
• Token budget management (what if the history is longer than the context window?)
• Logging, audit, and replay
• Handling new SDK versions when the provider ships a breaking change

What the same thing looks like on Nimblesite

curl -X POST https://api.nimblesite.dev/api/v1/chat/$CONFIG_ID \
  -H "X-API-Key: $TENANT_KEY" \
  -d '{"message": "Hello"}'

That's it. Memory, loop, tool dispatch, multi-tenancy, prompt templating, provider failover, logging, audit — all already done.

The honest comparison

When a raw LLM API is a better fit

• You're making one-shot completion calls with no memory and no tools
• You need absolute control over the exact bytes on the wire to the provider
• You're implementing novel agent architectures that don't fit a standard loop

When Nimblesite is a better fit

• Everything else.