Tracer and playground

Trace a conversation

Set trace_id in the config. MMSP saves the full history and the config after every response, before it yields the stop event.

for await (const event of client.streamingResponseStateful({
  message: { role: "user", content_items: [{ type: "text.done", text: "Hello" }] },
  config: { trace_id: "agent1/conversation_001" },
})) {
  // ...
}

async for event in client.streaming_response_stateful(
    message={"role": "user", "content_items": [{"type": "text.done", "text": "Hello"}]},
    config={"trace_id": "agent1/conversation_001"},
):
    pass

Traces are written under cache, or under the directory named by the environment variable MMSP_CACHE_DIR. The id above writes two files:

.json and .txt are appended to the id as it is, so v1.2 writes v1.2.json. An id that is empty, absolute, or leads outside the cache directory, through .. or a symlink, raises UnsupportedParameterError before the request is sent.

There is no database and nothing to set up.

Read the traces

Start the viewer and open http://127.0.0.1:25750.

import { Tracer } from "@prismshadow/mmsp/integration/tracer";

new Tracer().startWebServer("127.0.0.1", 25750);

from mmsp.integration.tracer import Tracer

Tracer().start_web_server(host="127.0.0.1", port=25750)

Or start it from a shell (the TypeScript scripts run in src_ts of a clone of the repository):

npm run tracer

python -m mmsp.integration.tracer --cache_dir ./cache --host 127.0.0.1 --port 25750

Playground

The playground is a local chat page for trying a model by hand: pick a model, set the config, send messages, watch the events. Picking a model fills in its client type and the endpoint that client reaches, the environment's or the vendor's own; change either to try the model through another protocol or a gateway. List models adds what the endpoint serves as entries that keep the client type and endpoint they were listed with, plus the latest key; listing again with only another key adds nothing. Sending to another model, client type or endpoint starts a new conversation under a divider; the messages above it are not sent; a key or extra-headers edit continues the conversation. Open http://127.0.0.1:25751; the tracer is served at /tracer/ next to it. The server page at /server/ starts an MMSP server from a table saved to a config file.

import { startPlaygroundServer } from "@prismshadow/mmsp/integration/playground";

startPlaygroundServer("127.0.0.1", 25751);

from mmsp.integration.playground import start_playground_server

start_playground_server(host="127.0.0.1", port=25751)

From a shell:

npm run playground

python -m mmsp.integration.playground --host 127.0.0.1 --port 25751