A HARNESS BY BLACKSQUID ENGINEERING
Figure it out
before you build it.
Your idea deserves more than another AI chat.
Bring the model you already use. Explore the questions and evidence on one connected board.
See what holds up. Leave with a plan you can build.
BRING YOUR OWN INTELLIGENCE · WE PROVIDE ITS ARTICULATION · PAID BY CREDITS
Building is easy now.
Knowing what to build is not.
Most ideas fail before the first line of code. Not because they were bad. Because nobody worked them out.
- 01
You open five chats and forget what the third one said.
Every chat is a straight line. Your thinking is not.
- 02
The AI writes ten pages. You read two.
The important sentence is in there somewhere.
- 03
You copy things into a doc, then another doc, then lose both.
Links, notes, screenshots. All over the place.
- 04
The AI talks you into something. You can’t trace back to what you wanted.
Where did this start? Why did I change my mind?
- 05
At the end you have a hundred pieces and no plan.
So you start building anyway. And find out the hard way.
You don’t need the answer.
Just a place to start.
Bring a messy idea. Work it out on one board.
Leave with one clear plan.
- 01 / THE QUESTION“What if…?”A hunch is enough.
- 02 / ONE BOARDWork it out.Nothing gets lost.
- 03 / ONE CLEAR PLANWrite it down once.Your build tool follows it.
- 04 / YOUR NEXT MOVESoftware. Hardware.Take it where you build.
A question you can keep working on.
Why is the café terrace empty when the sun comes out?
- On the canvas
- A paper and weather observations, with their connections visible.
- Still open
- Does shade change what people feel at the table?
An illustrative question, not a research result.
Review the scope first.
Mark the cards, notebook entries and answers that belong. A scope check can point out omissions; you choose what to include.
- Overview
- Problem, solution idea and open questions
- Build plan
- Modules, waves, PRDs and checks
- Full spec
- The complete Markdown definition
What you can take away Once a full package is composed, download an archive with PACKAGE.md, AGENTS.md and one PRD per module. Hardware files depend on the package.
Pull on a thread.
See what connects.
A paper. A weather feed. Something you noticed. Put them on the same canvas and start asking better questions.
Ask the AI about a selection, send an agent to research, or test a hunch against data. Keep the evidence that holds up. You decide what belongs.
Explore the canvas“Why is the café terrace empty when the sun comes out?”
TRY THE CANVASGet your hands on the canvas.
Move the cards, follow the agent’s thread, or change a temperature assumption in this interactive example.
A little less guessing.
A better place to sit.
A local comfort monitor for small cafés.
All that thinking.
One clear definition.
Select the evidence worth building on. Create a build package. The harness turns it into a bounded product definition with the context your next tool needs.
- 01
A reason to build.
The problem, the people and the evidence behind your idea.
- 02
A plan that fits together.
Architecture, modules, dependencies and a sequence for the build.
- 03
A way to know it works.
Testable requirements, acceptance criteria and the questions still open.
Look inside the package
- Product
- Problem, solution, concept, data and design system.
- Build
- Architecture, modules, build waves, PRDs and agent coordination.
- Validation
- Controls, tests, dependencies, risks and cited evidence.
- Hardware
- Bill of materials and geometry specifications, with generated drawings added to the export.
PACKAGE.MD · AGENTS.MD · CLAUDE.MD · PRD/ · BOM.CSV · CAD/
Ideas don’t all
live on a screen.
From “what if”
to “this fits here.”
Take your idea into physical form. Hardware packages define the parts, dimensions, materials and connections. Generate CAD drawings from the part specifications, then take the files into your engineering workflow.
- 01CAD source & drawingsShape + dimensions
- 02PCB specificationsSchematics + connections
- 03Bill of materialsParts + quantities
Drawing generation uses a connected runner. Available exports depend on the CAD engine; electronics output includes schematic source, SVG and a netlist.
Start with your hardware ideaComponents → connections → schematic.
Give your builder
the whole picture.
Take the architecture, requirements and acceptance criteria into the tool you already use. Your builder starts with the decisions you’ve made and the evidence behind them.
- 01VS Code + your agentDownload the archive
- 02Lovable / Base44Copy + paste the prompt
- 03Replit AgentCopy + paste the prompt
The archive includes AGENTS.md, CLAUDE.md and module PRDs. Web build tools open with the build prompt copied, ready for you to paste.
Start with your software ideapackage.md
agents.md
prd/dashboard.md
prd/weather-feed.md
prd/comfort-model.md
A little shade.
A lot more comfortable.
Choose the intelligence
for your work.
Connect OpenAI, Anthropic or Gemini with your API key, use a Claude Code subscription, or choose a qualified Synatrail-managed route when available. Choose the model source, set a limit and decide what it can work with.
With your own key or plan, your provider bills model usage. With qualified managed intelligence, model usage is charged in platform credits at the displayed price based on actual tokens. Either way, the harness adds context, tools, bounds, evidence and packaging that turn model answers into work you can check and build from. Main-action credit fees stay separate. What credits buy.
Choose model accessOne balance.
Clear charges for each source.
Your own thinking on the board is never metered. Credits pay the displayed fee for a job you commission. When qualified managed intelligence is chosen, the same balance also pays for separately priced model usage based on actual tokens.
See what credits buy- 01
Never metered
The canvas, your sources and data, the notebook, evidence review, accepting or rejecting the agent’s proposals, the community, and every export of your own work.
- 02
Choose your model source
With your own API key or Claude plan, the provider bills your model usage. A qualified Synatrail-managed route uses platform credits at the displayed model price based on actual token usage. Each managed call has a bounded hold and a usage receipt; unavailable routes cannot be selected.
- 03
Credits, per job
The displayed main-action fee is separate from managed model usage. One bounded research run, one researched pitch, one build release, one drawing job. You see the fee and a model spending limit before work starts. An unused action fee is restored under the job’s delivery rules.
YOUR TURN
What keeps you
wondering?
Bring that question. Give it a canvas. See what it becomes.
Built by Blacksquid Engineering for anyone who has to decide what to build. Read how it started or write to us.
Do I need an idea already?
No. A question or a passing thought is enough. The idea forms during the investigation.
What is the plan I leave with?
We call it a build package. One document made from your investigation: the problem, who it is for, what to build, how the parts fit, how to test it, and what is still open. For hardware it adds the parts list and drawings. It downloads as the files a build tool reads first.
What becomes public?
Nothing by default. You choose which sources, questions or interpretations to publish. Build packages stay in your project.
What does it cost?
You pay in credits for the articulation of each job you commission: a research run, a researched pitch, a build release, a hardware drawing job. The price shows before you start. Your own thinking on the board is never metered, and with your own key or plan your provider bills model usage; with Synatrail-managed access, separate model usage charges use the same credit balance. Buy credits in your workspace before you commission a job. See the pricing.
How do I learn it?
Follow one question through the documentation: gather sources, connect them, bring the agent in, run an experiment, share the result.