> ## Documentation Index
> Fetch the complete documentation index at: https://docs.rippletide.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Manual Python setup

> Instrument synchronous Python model, tool, and response call sites yourself.

Use this guide when you maintain the instrumentation yourself. First [install the Python SDK and create a connection](/docs/connect-python#install-and-connect). The generated setup prompt can adapt these same boundaries to your repository.

This integration supports Python 3.10+ agents with synchronous call sites. Python MCP servers, asynchronous execution and response streaming are not supported.

## Instrument your existing request handler

This fragment belongs inside your application. `SYSTEM_PROMPT`, `TOOLS`, `raw_model_call`, `run_existing_agent`, `trusted_context` and `send_to_caller` are your existing prompt, tool definitions, provider call, existing tool loop, authenticated context and response sink. Keep your real tool loop and add a guard at its dispatcher.

```python theme={null}
from rippletide import Rippletide, load_rippletide_env

load_rippletide_env("/absolute/path/to/your-agent/.env")
rippletide = Rippletide(name="Support agent")

rippletide.register_prompts({"system": {"fallback": SYSTEM_PROMPT}})
rippletide.tools("core", TOOLS)
rippletide.connection("provider", env={"api_key": "PROVIDER_API_KEY"})

def handle_request(request, trusted_context):
    def dispatch_tool(tool_name, tool_args):
        return rippletide.guard_tool_call(
            toolset="core", tool=tool_name, params=tool_args,
            context=trusted_context,
            execute=lambda evaluated: tool_handlers[tool_name](evaluated),
        )

    try:
        with rippletide.run(name="inbound_request", input=request):
            with rippletide.agent(name="orchestrator"):
                call_model = rippletide.traced(
                    raw_model_call, name="provider.request", kind="model"
                )
                candidate = run_existing_agent(
                    request, call_model=call_model, dispatch_tool=dispatch_tool
                )
                return rippletide.deliver_response(
                    request=request,
                    response=candidate,
                    context=trusted_context,
                    deliver=send_to_caller,
                )
    finally:
        rippletide.flush()
```

At the real tool dispatcher, call `guard_tool_call()` with the declared toolset/tool, parameters, trusted context and execution callback. Use the repository-specific generated prompt to fit that wrapper to your application's dispatcher. A policy check only protects paths that honor its outcome.

`load_rippletide_env()` does not override process environment variables. The SDK also accepts the app/manual `RIPPLETIDE_API_KEY` and `RIPPLETIDE_AGENT_ID` configuration. Never put a Platform key in the agent.

## Capture and failure behavior

Runtime capture defaults to **metadata**. Prompt text is withheld unless its owner explicitly enables `allow_content_export=True` on `register_prompts()`. Choose `RIPPLETIDE_CAPTURE_MODE=redacted` or `full` deliberately; see [Privacy & capture](/docs/privacy).

For tool and response guards, an enforced `BLOCK` raises `RippletidePolicyBlockedError` before the callback. Observe-only `WOULD_BLOCK` permits it. Network errors, 5xx, rate limits and malformed/interrupted HTTP responses fail open. Deterministic 400/401/403/404 decision errors raise `RippletidePolicyDecisionError`; do not treat them as telemetry outages.

For a declared result collection, `guard_tool_result()` requires explicit `allow_result_export=True`. The complete result is sent for transient evaluation, and your `apply` callback preserves the original response shape while delivering only the allowed items. See [Result filtering](/docs/result-filtering) for the contract and fallback behavior.

## Verify a real run

Run a turn that uses a guarded tool and delivers a response. Check `rippletide verify`, then `rippletide events --wait`. The app must show the inventory, heartbeat, tool and turn evidence, and the response-delivery outcome in **Traces**. Recording only a model candidate does not prove delivery.