This is the core guide to eight foundational workflow tools inside the fourteen-engine Project Telos roster. Each does a different job and leaves behind something you can check. For the complete Flywheel map, open the engine room. No jargon required.
If you only read one thing, read the next section. Everything else is the same idea, pointed at a different job.
Plate 01Overprinted ink rings from the same engine that draws every page here, turned up to full volume.
eight core tools · fourteen-engine roster · how to use each · where to get it
One idea, in three steps.
Most AI tools hand you an answer and ask you to trust it. These tools do something plainer instead. They look at real material, check it against a rule that is written down rather than felt, and hand you a small record you can re-open later. If you understand these three steps, you understand the shared core pattern.
1 · Perceive
Look at the real thing
A file, an image, a web page, a stack of sources, a claim. The tool works from the actual material, not a vague summary of it.
2 · Check
Against a written rule
Does the function only touch what it declared? Does every sentence trace to a source? Does the claim hold up when you try to break it? The rule is explicit, so the answer is not a matter of opinion.
3 · Keep the receipt
So it can be re-checked
What was seen, what changed, what stayed uncertain. A record a person can inspect after the answer has left the chat, and that reports one of three honest verdicts: it holds, it drifted, or it could not be verified.
Eight core tools, and what each is for.
Each tool points that same habit at a different job. You do not need all of them. Most people start with one that matches a problem they already have, then use the full fourteen-engine map when they need the larger Flywheel.
gatherresearch intake
Pulls in what you want to read (a video, a web page, a PDF, a paper, a feed) and records exactly where each piece came from, so the source survives long after you have read it.
Think of it as
A research assistant who staples a receipt to every source before handing it to you.
You use it when
You need a paper trail: which article, which timestamp, which transcript line a claim actually came from.
How to get it
pip install gather-engine
gather run <url-or-file>
Turns a big pile of code, or dozens of repositories, into a readable map of how things actually connect, so you (or a model) can see the shape of a project instead of guessing.
Think of it as
A floor plan for a building you only ever knew one room at a time.
You use it when
A codebase got too big to hold in your head, or you want to hand a model a compact, honest picture of it.
Runs a team of agents with routing and quality gates, then records who did what and why in a witnessed causal ledger that can be replayed step by step.
Think of it as
Meeting minutes for a team of assistants, with the route, decisions, and handoffs left open to inspect.
You use it when
A job needs several model calls or steps and you need the run to be accountable, not a black box.
Registers a thesis, steelmans its claims, measures them against a substrate, and refines the weakest axis until the result is useful, with the full paper trail included.
Think of it as
A devil’s advocate who shows their work, and admits when a claim is too vague to test.
You use it when
You want a claim stress-tested rather than rubber-stamped, with the reasoning left on the table.
A programming language where every function has to declare, up front, what it is allowed to reach out and touch (files, the screen, the network). The compiler enforces it: reach for something you never declared, and the program will not build.
Think of it as
A recipe that lists every ingredient it is allowed to use, in a kitchen that refuses to cook if the chef reaches for anything else.
You use it when
You are a developer and you want the compiler, not a code review, to catch a function quietly touching the network.
Turns material you already have into a course with a real integrity floor: spaced repetition, retrieval practice, and self-explanation checked by crucible. Every graded step stops and waits for you, and writes a witnessed receipt, so the record shows what you actually did.
Think of it as
A study plan built from your own notes, where nothing gets graded behind your back.
You use it when
You want AI help while you study something for real, with proof the practice was yours, not the model’s.
How to get it
git clone https://github.com/HarperZ9/learn
node src/cli.mjs doctor
A shared workbench with durable session state, workstation control, and sensory tools, where a person and a model can perceive and shape the same surface.
Think of it as
A shared workbench where both of you can see, discuss, and check every move, not a chat window where the model works out of sight.
You use it when
You want to actually collaborate with a model on a shared surface, rather than trade messages and hope.
How to get it
# public workbench
git clone https://github.com/HarperZ9/telos