Multi-provider
OpenAI, Gemini, Ollama, LM Studio, and any OpenAI-compatible endpoint.
Data Quixote
Test your LLM prompts on local data. Run it through multiple models at once. Compare outputs, check costs, export what works.
Load a local CSV, write prompts that reference any column, and run them through OpenAI, Gemini, Ollama, LM Studio, or a compatible API. The table, raw responses, token costs, and timing all stay in one window. Nothing leaves your machine.
Use hosted APIs, local model servers, or any OpenAI-compatible endpoint.
Provider Supported models
OpenAI gpt-5.4, gpt-4.1, gpt-4o
Gemini gemini-2.5-pro, gemini-2.5-flash
Ollama llama3.2, qwen2.5, mistral, gemma3
LM Studio Gemma, Llama, Qwen, local downloads
Compatible API Any exposed OpenAI-compatible model ID
Models
Model refresh
Pull model lists from providers that expose /v1/models.
Manual models
Add model IDs for local servers or gateways that cannot list models.
Side-by-side
Run two or three models on the same rows and compare their outputs column by column.
Data
File support
Open CSV, TSV, tab-delimited, JSON arrays, or Excel files.
Column variables
Reference any column with {{column_name}} inside a prompt.
One file, many prompts
Attach multiple prompt tabs to one dataset — test variants without re-importing.
Export
Download the completed table as CSV when the run finishes.
Prompts
Prompt variants
Compare different instructions against the same data.
System message
Write a system prompt once; every row in the run uses it.
Parameters
Tune temperature, top-p, max tokens, and reasoning level per model.
Run limits
Test all rows or sample the first 10, 100, or 1000 rows.
Runs
Concurrent processing
Process multiple rows at once instead of waiting for each response in sequence.
Pause and resume
Stop a run and continue later from the saved queue state.
Retry failed rows
Re-run only the outputs that errored.
Raw response viewer
Inspect and copy the model response behind any cell.
Stats
Token usage
Track input and output tokens by run, row, and model.
Cost estimate
See what each model is costing you, per row and in total.
Latency
Compare response time and throughput while a run is active.
Quality signals
Review similarity and ROUGE metrics when references are available.