Lab
Built in the open.
Folded into the work.
The lab is where the work gets tested first.
Some of it ships as open source. Some stays private. The point is simple: client work should be easier to trust, easier to run, and harder to break.
Tested under load
This is where workflow, memory, briefing, and QA ideas get used early, broken, and cleaned up before they reach client work.
Open when it can be
Public repos where that makes sense. Internal patterns where it does not.
Useful beats impressive
The goal is work that still holds together a month later, not a good-looking demo.
Why the lab matters
Where ideas earn their keep.
This is where new ideas get used early, broken, and tightened up.
Some of the best pieces are not public repos. They are the quiet parts underneath the work: state, memory, briefings, and anti-rot checks.
Open work matters because you should not have to trust a black box to get useful automation.
Four entry points
Four entry points.
Same working logic.
Client work does not inherit every repo. It gets the parts that hold up: execution, memory, briefings, and checks.
Workflow Systems
For repeat work, coordination, tool use, and handoff.
agency-cli
Command-line tools for repeatable creative work, multi-agent coordination, and fewer one-off handoffs.
phantom-cli
Content pipeline tooling that cuts down manual glue work across review, generation, and publishing.
scty-mcp
The plumbing that lets models use the right tools instead of hallucinating around them.
shared state
A shared record of runs, deliveries, and events so work does not disappear into chats and terminal scrollback.
Living Knowledge Systems
For turning raw source material into memory a team can reuse.
wiki
A maintained memory layer for claims, programs, voice, and the facts that need to stay straight over time.
tokdown
Markdown tooling for keeping long source material structured, readable, and model-friendly.
pdf-cli
Pulls usable text and structure out of PDFs so research does not die in attachments.
civic-cli
Research tooling for policy and legislation when the answer has to point back to real sources.
Briefing Systems
For daily synthesis, leadership updates, and decision support.
nightly loop
A scheduled loop that gathers activity, resolves loose ends, and turns the day into a usable brief.
persona briefs
Briefs that keep priorities, continuity, and relationship context intact from week to week.
brandOS
Interfaces for seeing search and model signals side by side instead of guessing what the machines see.
field reports
Public essays that turn research and field notes into something sharper than a recap.
Assurance Systems
For monitoring, evals, and checks that catch drift before anyone else does.
tripwire
Checks for freshness, delivery failures, source health, and the quiet breaks that rot a workflow over time.
autoresearch
An experiment loop that tests changes, measures the result, and keeps the ones worth keeping.
critbench
Benchmarking for reasoning quality when “looks smart” is not enough.
givecare-bench
Benchmarking for high-stakes AI where tone, memory, and judgment all matter.
Demos and field notes
Working ideas, shown in public.
Pretext Demo
Interactive text layout without the DOM. Small, strange, and a reminder that the lab still makes room for experiments.
AI Surfaces: A Field Report from CYLNDR Off-Site 2026
A working point of view on where AI actually is right now and what operators should pay attention to.
Your Next Customer Isn’t Human
What agent-mediated discovery changes for brands, products, and the surfaces around them.