Build better AI workflows.
Guides, experiments & open tools
Keep useful context. Find the evidence. Pick the tool that helps with the work in front of you.
LLMSX is for developers and technical users working with AI agents. Start with a worked example, inspect the result, then reuse the instructions or reference files in your own setup.
Try your first workflowRead the guides & research
Read the guides and download reference files without an account.
Original experiment · 256 agent runs
Do better docs actually save an AI agent money?
On two small repositories, llms files and search indexes on disk did not produce measurable savings. Loading documentation into the agent’s context did. That distinction changes what is worth building.
The write-up includes the questions, raw results, correctness checks, failed setups and limits of the experiment.
Read the experiment →Solve a concrete problem
All guides & research →Architecture guide
Resume work without replaying everything
Separate a small handoff from a durable recall store. Includes design choices and the measurements still needed.
Implementation report
Find the exact flag your agent missed
See how keyword and hybrid retrieval handle literal tokens, with queries, failure cases and commands.
Repair report
Check whether your citations actually resolve
Follow a broken-anchor investigation from the extraction bug to a reproducible validation step.
Choose a tool for your task
Requirements & installation →Three separate terminal workbenches. Choose the one that matches the job; you do not need the whole stack.
- Get LLMSX →
LLMSX Explorer
Browse the concept library, keep notes and organize research jobs in a terminal workbench.
- Get Skills Explorer →
Skills Explorer
Inspect, edit and maintain the skills already installed across your agent harnesses.
- Get GitHub Explorer →
GitHub Explorer
Browse a repository, edit with Vim and inspect Git and GitHub commands before running them.
Want instructions instead of an app? Browse 23 reusable agent skills and read each full prompt before installing it.
Connected reference library
Explore the concept tree.
Browse a topic, inspect its published reference and follow its sources. Download the facts as Markdown when you want to use them with an agent.
The library includes agent-assisted research and mirrored source reports. Source links help you check a claim; they do not certify it.
- Download context filesResearch reports and facts, sized for loading · Markdown index
- Inspect the source directoryllms files scored for format quality
- Read the technical referenceFormats, lint rules and publishing conventions
- Use the code examplesCopyable recipes with stated assumptions
- Inspect the retrieval demoA recorded keyword, vector and hybrid comparison
- Load the site through llms.txtAn index for agents, with links to the published files
An open project by Mitch Hudson. Read how to interpret the sources and results, or report a correction.