Skip to main content

Chat with Your Traffic

proxymock web has an assistant built in. Ask it what a recording contains, why a request failed or which blueprint would fix a mock miss, and it reads the RRPair files in your workspace to answer. It can also act for you: configure and start a replay, step through a replay debug session, create and apply blueprints, and push or pull cloud snapshots. The actions that require approval are listed below.

Chat needs a Speedscale enterprise account on the machine running proxymock web. Without one you can open the assistant, but sending a message shows where to get a pro account.

Open the assistant​

Start proxymock web:

proxymock web

On an empty workspace, click Ask the assistant to record traffic for your service. The floating Chat button in the bottom-right corner is hidden by default. To show it in a workspace that already has traffic, run localStorage.setItem('pmShowChatButton', '1') in the browser console and reload the page.

The assistant opens with suggestions such as Summarize this traffic, Explain this request and Find similar RRPairs. Press Enter to send, and Shift+Enter for a new line. While it answers, Send turns into Stop.

The assistant knows which view you are on and what you have selected, so "explain this request" means the RRPair open in front of you. To point it at something else, type @ and pick a file or directory from the workspace. Mentioning a directory such as a recording or a replay results run lets it work across every RRPair inside.

The header has Copy to copy the conversation, Clear to start over, and Reset to put the panel back where it started. The conversation lives in the browser tab. It survives switching views but not a page reload, and each tab has its own.

What it can do​

Without asking first, the assistant can:

  • List, search and read the RRPairs, runs, blueprints, transforms and dataframes in the workspace.
  • Read a replay's status, report and logs.
  • Run the analysis and list recommendations, including a dry run of what applying them would change.
  • Read your Speedscale cloud data: clusters, workloads, snapshots, reports and their recommendations.
  • Open an RRPair, the Blueprints view or the Requests view for you.
  • Fill in the Replay form, except for the wrapped application command and output directory, finish an open replay debug session, or set a variable in it.

These actions show an Approve card first, and nothing happens until you click it:

  • Set the Replay form's wrapped application command (user_command) or output directory (out_dir).
  • Start a replay, or start a replay debug session, step it, run it to a breakpoint, or resend the last request.
  • Create, activate, apply or import a blueprint, overwrite one from the cloud, or sync the workspace's blueprints with Speedscale cloud, which pushes local-only blueprints up.
  • Apply recommendations to RRPair files. proxymock backs up the files it changes under .proxymock/backups/ in the workspace.
  • Update or delete a dataframe.
  • Delete a run.
  • Push the workspace to Speedscale cloud as a snapshot, or pull a cloud snapshot into the workspace, which overwrites it.

While an approval is open, the assistant waits and you cannot send another message. Cancel tells it you declined.

Choose the model​

By default the assistant uses Speedscale's managed model. Requests go through your Speedscale login, Speedscale chooses the model, and you need no other key.

To use your own OpenAI or Anthropic account instead, pass the provider, a key and a model id:

proxymock web --chat-provider anthropic --chat-api-key "$ANTHROPIC_API_KEY" --chat-model <model-id>
proxymock web --chat-provider openai --chat-api-key "$OPENAI_KEY" --chat-model <model-id>

--chat-model is required, since there is no default model for your own provider. When --chat-api-key is left out, proxymock reads SPEEDSCALE_CHAT_API_KEY, and for Anthropic then ANTHROPIC_API_KEY. An Anthropic key that starts with sk-ant-oat, such as one from claude setup-token, is sent as an OAuth token. Requests go to the provider's public API; a custom endpoint, such as a local model server, cannot be configured.

Where your data goes​

To answer, the assistant reads RRPairs, including headers and bodies, and sends what it reads to the model, along with a short description of the current view. Tool results are capped at 25 KB each. Nothing is redacted on the way to the model, so if a recording holds secrets or personal data, redact it first with DLP rules and chat over the redacted copy.

Managed modelYour own provider
Model requests go toSpeedscale, which passes them to its model providerYour OpenAI or Anthropic account, directly
Conversation stored by SpeedscaleYesNo
KeyYour Speedscale loginYours, sent only to the provider

With your own provider, the panel shows a Private badge. Questions about your cloud data still call the Speedscale API with your login in both modes.

Turn it off​

Start proxymock web --chat=false to hide the assistant entirely. The chat flags are then rejected, so a script cannot turn it on by accident.