Open source from Trysti Labs
Alongside client work, we build and release open source tools.
The research and infrastructure we use to ship reliable AI, given back to the community.
OptArena
OptArena is the reproducible evaluation and regression platform for AI coding agents. Run CLI agents, in-process SDK and agent-framework agents, and raw-model baselines through the same tasks against any OpenAI or Ollama-compatible backend, with every verdict coming from compiling and running the generated code against hidden tests. Replace opinions with evidence, and know for certain when your agents actually get better.
Compare every agent
The same task through CLI tools, in-process SDK frameworks, and raw-model baselines, on identical inputs and one shared oracle.
Verify by execution
Generated code is compiled, run, and checked against hidden test suites inside an isolated container. Proof, not keyword-matching.
Catch regressions
Diff any run against a baseline: accuracy, cost, and runtime deltas, plus exactly which cases broke, with a non-zero exit to gate CI.