AI7 min readJuly 15, 2026

ASK AI: Query Your Load Tests in Plain Language

Set up the built-in AI assistant with your own provider, then ask questions about runs, metrics and machines — and get live tables, charts and action buttons back.

By Perfscale Team

What you'll set up

By the end of this guide the AI tab in your perfscale dashboard will answer questions like "what was the highest request rate in the last 3 runs?" with live tables, charts and one-click links into your workspace.

You'll need:

  • A perfscale workspace where you are owner or admin (any plan)
  • An API key for any OpenAI-compatible or Anthropic-compatible LLM endpoint

Step 1 — Connect your provider

Open Settings → AI provider:

  1. Provider type — pick the wire format your endpoint speaks:
    • OpenAI compatible (/chat/completions): OpenAI, OpenRouter, DeepSeek, Groq, vLLM, Ollama behind a proxy…
    • Anthropic compatible (Messages API): Anthropic, or any gateway speaking that format
  2. Base URL — e.g. https://api.openai.com/v1 or https://openrouter.ai/api/v1. Pasting the full endpoint path (…/chat/completions) also works — we normalize it.
  3. Model — e.g. gpt-4o, claude-opus-4-8, deepseek-chat.
  4. API key — stored encrypted (AES-256-GCM), shown afterwards only as its last 4 characters.
  5. Optional extra headers for gateways that need them.

Hit Test connection — it runs a one-token completion against your candidate config before saving, so typos surface immediately.

Privacy note: your key and your data never pass through perfscale-operated AI. The model you configured is the only LLM in the loop.

Step 2 — Ask

Open the new AI tab (visible to your whole team once configured) and ask in any language:

  • Which run had the worst p99 this week?
  • Show the last 5 failed runs as a table
  • Plot http_req_rate for the latest run
  • Which machines are offline right now?

The assistant answers with real dashboard components — tables, line charts, stat cards — and every run answer carries buttons to open the run page, jump to metrics, or download the raw log.

Under the hood the model calls read-only workspace tools (runs, tests, machines, metric catalog and time-series queries) using your access token, so it can never see more than you can.

Step 3 — Share and revisit

Every conversation is a persistent thread with its own URL:

https://your-perfscale/admin/ai/chat/<thread-id>

Threads are private to each user; the sidebar keeps your history.

Limits

The LLM tokens are billed by your provider. perfscale meters only the workspace data-tool calls the chat makes:

PlanTool calls / month
Starter100
Growing1,000
Enterpriseunlimited

Chat history counts toward your plan's storage quota alongside run logs.

Troubleshooting

  • "Test connection" fails with Forbidden — some providers geo-block server IPs; try a provider reachable from your deployment region, or route through your own gateway.
  • "AI provider is not configured" in chat — the provider is set per-workspace; switch to the workspace where you configured it.
  • Limit reached — the chat tells you the exact usage; upgrade the plan or wait for the next calendar month.

Comments

Reply on Bluesky