stevibe

BenchLocal is a local-first desktop app for running practical benchmark packs against local or remote models. Point it at Ollama, LM Studio, OpenRouter, or any OpenAI-compatible API, install a Bench Pack, and run real scenarios for extraction, tool use, debugging, structured output, and agent workflows.

Built for deterministic scoring, verifier-backed benchmarks, and comparing models without giving up local control.

Ever wanted to ask Claude, GPT, and Gemini the same question in the same conversation? PingModel is a multi-bot workspace where each model gets a handle, a system prompt, and a seat at the table. @mention who you want, keep the conversation in one place, and let them argue it out.

Built for comparing perspectives, not tab-switching.

Capture, inspect, and replay HTTP traffic on your phone. Built for the developers who instinctively hit F12 — even on mobile. Browse a page, watch the network calls fly, inspect the payloads, then send modified requests through the sandbox.

Because sometimes the bug is in the API response, and you're not at your desk.

I needed to pass the IELTS. Existing tools felt like homework. So I built a gym instead — pick a difficulty, do your reps, get real-time feedback from an AI coach on grammar, word choice, and structure. Turns out other people needed it too.