A question worth asking.
Your knowledge of the problem matters. We begin with what you want to change, who it should help, and what a useful result would look like.
AN OPEN WORKSHOP FOR PEOPLE + AI
Bring your
curiosity.
See what follows.
What could a small team build if it had a little more help thinking, exploring, and making?
Meet Team6. Six AI roles, working together under your direction. Built to help turn ambitious questions into work you can use.
Step insideNo special vocabulary required. Just a question.
HUMAN PURPOSE. MANY WAYS FORWARD.FIG. 01
01 / THE POSSIBILITIES
Start with a
“what if?”
You don’t need to know which model to use.
Start with something you would like to make possible.
ILLUSTRATIVE WALKTHROUGH 01 / 03
From “I have an idea”
to “try it for yourself.”
Imagine a local repair shop that needs a clear, useful website. The team explores what customers need, plans the pages, designs the experience, builds it, and checks that the important parts work.
WHAT YOU WOULD REVIEW
A working preview, the decisions behind it, and any issues still to resolve. You decide what goes live.
What makes that work?
Each role passes an actual piece of work to the next: research, a plan, a design, then code. A separate reviewer checks the result. The handoff includes what is finished and what is still uncertain.
An explanation of the workflow, not an AI session running in your browser. Results depend on the models, setup, and human review.
02 / A WINDOW INTO THE WORK
A little less mystery.
A lot more possibility.
Here is what coordination looks like behind the scenes. Work is divided, knowledge is carried forward, and results are checked before they reach you.
team6 · inside the workshop
ILLUSTRATIVE REPLAY · narrative steps, not a live process
→ Intake: Lugia receives your brief and clarifies the goal.
→ Frame: agree on the outcome, constraints, and review boundary.
→ Route: choose the specialist role the work actually needs.
→ Knowledge: load relevant decisions and the sources behind them.
→ Local helpers: Redact · Gist · Title; hold uncertain results.
→ Models: use local or hosted models to suit the task and setup.
→ Handoff: save the artifact, owner, verdict, and open questions.
→ Pipeline: frame → research → architect → design → build → verify → report.
$ python3 build/sweep-gate.py # kit assembly: check the sources
$ python3 build/review-gate.py # require the review record
$ python3 build/generate.py --params examples/demo-consulting.yaml --out /tmp/demo-team
→ Review: check the delivered result, not a claim that it worked.
→ Report: show what is ready, what is uncertain, and what comes next.
TEAM6 / 6 AGENTS · READY FOR HUMAN REVIEWReal build commands sit alongside explanatory steps.
See the runnable setup guide ↗
03 / THE HUMAN PART
You bring the reason.
We help with the how.
AI can explore, draft, and build. It can also miss the point. That is why Team6 gives the work a structure, and keeps people in charge of its purpose.
Different jobs. Shared notes.
A researcher gathers evidence. An architect makes a plan. A designer shapes the experience. A builder makes it. Each leaves something the next can use.
Someone checks the work.
A separate reviewer looks for mistakes. The director brings the results together. You see the evidence and decide whether the work is ready.
Director→Researcher→Architect→Designer→Builder→Reviewer→Director
Six roles. One shared project. Human judgment at the beginning and the end.
04 / KEEP ASKING
The interesting part
is how it works.
Open a question. Follow it as far as you like.
Can some of this stay on my own hardware?+
Yes, with a compatible local model and enough memory to run it. Small specialist models can do narrow jobs, while larger models handle harder work. Hosted models are another option. The right mix depends on the task, hardware, and privacy needs.
Team6 documents optional local helpers: Redact flags sensitive text, Gist suggests a topic, and Title drafts a label. These helpers do not replace reasoning or human review. An uncertain redaction must be reviewed before text leaves the device.
Explore local preprocessing ↗How does a project remember anything?+
Think of a workshop notebook. Decisions, sources, and completed work are saved in files. When a new session begins, the team loads the relevant notes instead of relying on a conversation that may be gone.
This takes explicit recording and sensible retrieval. It is a way to preserve continuity, not a promise of perfect memory.
Read about durable state ↗What happens when an agent gets something wrong?+
The work should stop at a check, return to the role responsible, and be corrected. If an agent stalls, bounded recovery rules limit retries and bring unresolved problems back to the person supervising.
Review reduces the chance of error. It does not eliminate it. Important decisions still need qualified human judgment.
See the working rules ↗Is this something I can actually use today?+
The open-source kit, templates, build gates, and generator are available now. You will need Hermes, Python, and some comfort with a terminal. Setup includes manual steps.
The conversational setup agent and a one-command installer are not shipped. The public kit has not yet been independently demonstrated to deliver a real product end to end. We keep that distinction visible.
Open the setup guide ↗THE DOOR IS OPEN
What have you been
wanting to make?
Explore the kit yourself, or bring us a question.
You don’t need to have the whole answer yet.
Available now Open-source kit · templates · generator · working rulesStill being built Conversational setup agent · one-command installer