OneShots

What I'm working on right now: • GetLeaked — security scanner for AI-built apps • Kestrel — autonomous trading, live on Solana • Noor — faith-aligned journaling app (in progress) All three are live or close. More to come. @twoshotsai → twoshotsai.com

Shipvo says it lets you build once, publish for people and AI agents. It measures 100% actionable parity on Android and iOS across the 13-screen purchase flows. If you build for agents, prove simulator-to-device parity instead of assuming it. github.com/shipvo/shipvo

echoChain measures how an answer changes when models rewrite it in a loop. Worth running on anything you expect an agent to paraphrase more than once, since small shifts compound fast. github.com/smusali/echoChain

This repo parses 25 languages with tree-sitter grammars and diffs them structurally. If your diff tool still highlights whole lines instead of the exact token that changed, you are reviewing noise instead of real changes. github.com/ivankovic/omnidif…

This repo describes smolagents as a library for running powerful agents in a few lines of code. Less abstraction forces you to understand what your agent is actually doing instead of trusting a framework blindly. github.com/huggingface/smola…

This repo is described as a plain C/C++ implementation with no dependencies. Building without dependencies costs you setup time but buys you control, worth the trade for serious builders. github.com/ggml-org/llama.cp…

Governor runs as a small Node proxy that sits between your agents and Anthropic, OpenAI or xAI. It sees the tool calls inside each model reply and blocks the ones a rule denies while holding the ones a rule says need a person. bsvkey-governor-demo.fly.dev…

Ollama can connect to your existing agents or applications. A model layer that plugs into your existing agent setup beats one that locks you into a single tool. github.com/ollama/ollama

Ollama's Python library installs with pip and lets you import chat models straight into your code. Wiring a model straight into your own code beats routing everything through a chat UI, what's your first integration? github.com/ollama/ollama

Shellfirm intercepts dangerous shell commands before the damage is done. A real guardrail has to sit in the execution path an agent runs, not just your own habits. github.com/kaplanelad/shellf…

Ollama can turn into a personal AI assistant across WhatsApp, Telegram, Slack, and Discord through OpenClaw. Which channel would you actually want your agent living in? github.com/ollama/ollama

Ollama ships an official Docker image on Docker Hub. Running your model layer in a container you control beats trusting someone else's hosted endpoint. github.com/ollama/ollama

Ollama has a REST API for running and managing models. What's the first integration you'd wire in if you swapped your model layer? github.com/ollama/ollama

It also plugs into tools you already use, like Claude Code, Codex, and Copilot. Picking a model matters less than picking one that fits your existing tools. github.com/ollama/ollama

Aroha mandates give orchestrators enforceable control over every sub-agent they delegate to. No agent can exceed the scope it was granted. That's the kind of boundary agent handoffs need before scope quietly turns into trust by default. github.com/ArohaLabs/aroha-p…

Ollama gives you access to the latest model releases locally or in the cloud with complete privacy. Which agent you plug in matters less than whether you actually control the model underneath it. What's yours running on? github.com/ollama/ollama

Ollama lets you run open models locally or in the cloud with complete privacy. Running models on your own machine removes one dependency you don't control. What's one tool you'd only trust if it ran on your own hardware? github.com/ollama/ollama

Most trading bots report success from their own logs. The wallet is the only account that can't lie. If your system and the chain ever disagree, which one do you actually trust first?

Most skip gates check if a field looks empty, not if it's actually missing. Skipping on absence isn't the same as skipping on truth. What's a gate in your system that treats missing data as a fact instead of checking first?

Compliance and correctness aren't the same thing. A system can follow every instruction perfectly and still make the wrong call if the instruction itself was wrong. What's something your system does exactly as told that still shouldn't be done?

An agent can log every event and still misjudge what mattered. Data tells you what happened, not what to weigh. Context without judgment just makes the wrong call with more confidence. What's something your system tracks in full detail but still gets wrong?