Hyperparam Quick Start Guide

Get started with Hyperparam in just 3 minutes. This guide walks you through loading agent traces or chat logs and running your first analysis.

Step 1: Sign In

Navigate to hyperparam.app and sign in with Google. Signing in unlocks AI-powered analysis, workspaces, and export.

Hyperparam Home

Step 2: Load Your Data

Upload a File

Drag a .parquet, .csv, or .jsonl file onto the drop zone (Claude Code transcripts, Codex sessions, ChatGPT exports, Langfuse / LangSmith / Phoenix traces, or any structured log file). Your data uploads to your Hyperparam storage and is ready to explore immediately.

Connect to Cloud Storage

Paste a URL to data stored in S3, GCS, Azure Blob, or any public endpoint. Hyperparam streams the data directly, no download required. See CORS Configuration if you're connecting to a private bucket.

Don't have logs yet? See Exporting Chat Logs for guides on pulling logs from Claude Code, ChatGPT, Langfuse, LangSmith, Phoenix, Datadog, and more.

Step 3: Explore Your Data

Once loaded, you'll see the interactive table view:

  • Scroll infinitely through millions of rows
  • Click column headers to sort
  • Double-click cells to inspect nested data (conversations, tool calls, JSON)
  • Drag columns to reorder

Step 4: Analyze in a Workspace

Click "Edit data in workspace" to open your data in a workspace. From here you can:

  • Ask questions in chat, e.g. "Which conversations had failed tool calls?"
  • Add AI-generated columns: failure classification, quality scores, root-cause categories
  • Create SQL views: filter, join, and project across your data
  • Export results: save filtered datasets as Parquet, CSV, or JSONL

Step 5: Try AI-Powered Analysis

Try these prompts in the workspace chat:

"Which sessions had failed tool calls? Add a column."
"Classify each failure as avoidable (prompt or schema error) or necessary (probing the environment)."
"Find conversations where the agent got stuck in a loop or burned tokens without progressing."
"Score each chat 0-1 for user frustration based on the user's messages."

Your First Workflow

Here's a workflow to try with your own agent or chat logs:

  1. Load your logs (or try an example dataset)
  2. Sign in to unlock AI features
  3. Ask chat to surface failed tool calls or agent errors
  4. Add a column classifying each failure as avoidable or necessary
  5. Create a SQL view filtering to avoidable failures
  6. Export the filtered dataset, or save the analysis as a skill to re-run next week

What's Next?

Need Help?

Hyperparam Quick Start Guide - Hyperparam