Its tracing covers application functions, HTTP requests, database queries, model calls and tool calls. The company provides syntropylabs-evalkit packages for Python and TypeScript; Go and Java use manual spans.
The service can ingest sessions from Claude Code, Codex, Gemini CLI, Cursor, Windsurf and OpenCode as traces, including token and cost data where a vendor reports it. It also flags errors, empty completions, truncation, tool loops and model swaps. Evaluation runs can use datasets, managed or custom prompt evaluators, pass-rate summaries and deployment gates. Simulations generate or run scenarios and record turns, tool calls, scores and verdicts.
Tracing is free for one project, with seven-day retention and a soft cap of 10,000 spans per month. Paid plans cover evaluations, datasets and simulations; enterprise pricing is listed as custom. The source does not establish paid-plan prices, independent performance results, customer adoption, or the accuracy of its evaluators and simulation outcomes.