Guth Labs · Experiment workbench
What we’re
working on.
Instruments, experiments, memory systems, research, and the lives of agents. Labs is where we’ll share what we’re making, how we test it, and what we find.
01 / Coming soon
Instruments
Tools for measuring, checking, and understanding a problem. Explore what each instrument does and the evidence behind its results.
Browse instrument documentation02 / Open workbench
Experiments
Five synthetic experiment packs for memory meaning, corrections, receipts, selective repair, and field guides. Compare local models, inspect results, and repeat runs with a Cloudflare coordinator.
Research notebook · In development
A record you can follow, with a researcher beside you.
The planned Labs workspace brings experiment questions, methods, model choices, runs, results, and sources into your research notebook.
A dedicated researcher you can name will help explain findings, design tests, and contribute to agent builds. Keep proposed improvements beside the evidence for them, including failed runs and open questions.
Compare agent versions and suitable local or cloud models, trace the components behind a build, and keep the next experiment connected to what you learned. These Labs research controls are in development.
03 / Coming soon
Memory systems
How agents remember useful information, find it again, and correct it when things change. This is where we’ll share the research behind that work.
04 / Coming soon
Research papers
Longer studies of agents, memory, local models, and assurance, with methods and sources you can follow.
05 / Coming soon
Publications
Reports, articles, technical notes, and updates from the lab, collected in one place.
06 / Coming soon
Games & playgrounds
Interactive games and challenges for exploring how agents behave. A place to try ideas and see what happens.
07 / Coming soon
Agent lineage
Trace how an agent develops: the versions, components, and changes that shaped it over time.
08 / Coming soon
Agent birthing
How a new agent takes shape: its purpose, identity, memory, skills, and first tests. We’ll make the creation process easier to understand.
09 / Coming soon
Inspections
Look inside an agent’s work. Follow the sources, permissions, actions, and receipts to understand what happened.
10 / Coming soon
Local models
Making useful AI work on your phone and computer. We’ll share our local model experiments and what we learn from them.
11 / Coming soon
Agent internet
How agents find websites, documentation, and tools. We’re exploring ways to make that information easier to discover and check.
12 / Templates & workflows
Give work a clear starting point.
Templates collect the information a project needs. Workflows explain the steps, responsibilities, and checks that follow.
Start with a template, fill in the requirements, and adapt the steps to the job. The documentation shows the structure for projects and agent work.
13 / Field guides
Know how to approach the job.
Field guides explain a method: when to use it, what information to gather, and how to check the result.
Use them alongside a task to follow a consistent process. Each guide provides a reference for the work and the boundaries that matter.
Browse field guides14 / Tools
Understand what a tool does.
A tool gives an agent a specific way to work. Its reference explains the job, required inputs, expected outputs, and limits.
Check the tool’s purpose and requirements before adding it to a workflow. The catalogue includes its documentation and qualification notes.
Browse tool references