I build developer tooling for the AI era — CLIs, profilers, and benchmarks that make working with LLMs less painful and more predictable. Vibe coder using Claude Code and OpenAI Codex.

Two lines of code, zero config. Wraps your OpenAI or Anthropic client and gives you instant visibility into spend by feature, hotspots by file, retry waste, context bloat, and model downgrade opportunities. Found $1,240/month in waste on my own project in the first run.

from llm_cost_profiler import wrap
from openai import OpenAI

client = wrap(OpenAI())  # that's it

pip install llm-spend-profiler · Python · MIT

CLI that sends the same prompt to OpenAI, Anthropic, Gemini, and Groq in parallel, streams a live terminal race, scores on speed/cost/quality, and generates shareable result cards.

llm-bench run "Explain vector databases" --models gpt-4o,claude-sonnet-4-6,gemini-2.5-flash

npm install -g llm-bench · TypeScript · MIT

CLI-first stability analyzer. Runs your prompt N times, scores pairwise similarity, gives you a stability label (High / Medium / Low), inline diffs, and a local JSON artifact. Published on PyPI.

ai-stability run prompt.txt --n 5 --provider openai --model gpt-4.1-mini

pipx install ai-stability · Python · MIT · PyPI

Python TypeScript Claude Code OpenAI Codex OpenAI API Anthropic API Gemini Groq CLI tooling SQLite Ink/React

• LLM observability and cost tooling
• Developer-facing AI infrastructure
• Fast, zero-dependency CLI tools
• Prompt reliability and evals

Open to collaborations, contract work, and conversations about LLM tooling. If you're building in this space, let's talk.

• X/Twitter: @BuildWithAbid
• GitHub: github.com/BuildWithAbid