Open source · Design team in the AI you already use

Your design team. Type the name.

Named specialists live in Cursor, Claude, or a browser tool. Type /zara. You get a verdict — not a polite paragraph. The design team is open source. No new app.

Type /zara /accept yes /share
See what you get It's open source

Then pick who you are · Full design team page

Open source. Read every file. A verdict, not a nice paragraph. You still decide every change.
1named team in the AI you already use
SHIPrevise or block — a verdict, not a paragraph
0silent changes. You keep or skip.
3jobs you can try without a designer
MITthe design team is open source. Read every file.
What you get

Three jobs. Human words.

You do not need a designer on staff. You do not need to know the slash name yet. The team stops at a human decision.

Before you build

Two sketches first

Two competing layouts before anyone opens Figma or writes code.

/ux-ideator
After it shipped

A verdict on the screen

Ship, revise, or block. Then keep or skip the note.

/zara /accept
Against the story

Can a person finish?

Can a real person complete the job using your PRD — not vibes?

/ux-story-gate

/zara /accept /share /receipt · All jobs on the design team page

Your path

Built for designers — and the people who sit next to them.

You know the product. Now pick who you are. Nothing here asks you to add a package to your app.

Dev / PM

I ship product

No designer on staff. Two sketches, a verdict, or a check against the PRD.

Start those jobs
Firm / owner

I run a practice

Same idea for books or visit notes. Packets arrive in the tools you already open.

See a practice packet
Same idea for a practice

A packet that reaches the desk ready to sign.

If you run a firm — not a product team — we map one repeat workflow. Books, drawings, visit notes, or POs. You still review. You still sign.

Lab products

Each product is a focused team.

Analyzthis Lab = shipped team products + custom workflow teams. Same method. Different delivery.

Pick what is live, what is still in the lab, and what is a mapped service. Do not treat the three as one SKU.

analyzthis_design

Design team

Live · open source · v2.5.2

A task-grounded design team for ideation, critique, deliberation, verification, and human-reviewed evolution. Open source — read every file. For designers: type /noor, then /accept and /share — no terminal flags.

You get
Commands in your IDE, verdicts, and the source on GitHub
You need
Cursor, Claude, Windsurf — or nothing at all if you work in Lovable, v0 or Bolt
What's new
/accept and /share — keep, skip, or send a correction without CLI flags
Start as a designer Read the source
analyzthis_product

Product workflow team

Lab / early access

Specialist roles for product discovery, PRD shaping, roadmap pressure-testing, and delivery-quality review.

You get
PRD and roadmap review once the roster ships
You need
A waitlist conversation — not a download today
Ask about early access
Custom teams

Built for your workflow

Service

We map one repeat workflow and design the smallest useful team around it. Packets arrive in the tools you already open.

You get
A mapped roster and packets beside your current tools
You need
Discovery first. No IDE required.
Map one workflow
The problem

Generic AI gives you generic work.

A single assistant has no durable division of responsibility, no internal challenge, and no definition of “good” that belongs to your profession.

01

It forgets context

Your standards, source material, and prior decisions are repeatedly pasted—or simply lost.

02

It agrees too easily

One model produces and approves its own work. Weak assumptions survive because nobody is assigned to object.

03

It never owns quality

There is no role-based gate, outcome history, or structured learning loop tied to your definition of done.

How we measure success

We score the handoff, not the chat.

A partner, doctor, or owner was going to review the work anyway. That review is the measurement. We do not ask the firm to fill a weekly scorecard.

Opened

Was it opened at all?

If the packet is never opened, the team is already dead. Abandonment is the churn signal nobody tracks.

Changed

How much did they change?

Accepted as-is, light edit, heavy rewrite, or discarded. That is first-pass quality, and it needs no survey.

Review time

How long did review take?

Not how long the work took to prepare — how long the human spent. That is the number that converts to money.

Repeat

Did it run again?

Monthly filing, weekly submission, daily list. If cycle four still runs, it works.

Team contract / exampleSigned at discovery
WorkflowMonthly GST filing prep
OwnsBooks organisation, missing-doc chase, packet assembly
Does not ownFiling, client advice, sign-off
Good looks likePartner signs in one review. No document found missing after review.
Fails whenA missing invoice is discovered post-review, or the filing date slips
Reviewed byPartner, monthly
We trackOpened / edit level / review minutes / repeat run
Weeks 1–3 are calibration. The team is tuned, not scored.

We do not grade the team in its first weeks. Everything looks rough, the firm loses nerve, and the account dies before the loop closes. Scoring starts after calibration.

How the packet is made04 specialists
01
ArchivistReads your knowledge and decisions
GROUND
02
PlannerBreaks work into accountable steps
PLAN
03
ChallengerFinds weak assumptions before delivery
DEBATE
04
VerifierChecks output against your standard
GATE
Shared context → one accountable outcome
Our method

The lab architecture behind every product.

Whether we are shipping analyzthis_design, building analyzthis_product, or designing a custom team for your workflow, the method stays the same: encode accountability around the task.

Map one workflow
01
Discover

Map the real workflow

We identify the repeated task, inputs, expert decisions, tools, failure points, and the outcome that matters.

02
Architect

Assign specialist roles

Each agent gets one allowed job, explicit boundaries, the right context, and a handoff contract—no vague “do everything” prompts.

03
Ground

Connect your knowledge

PRDs, SOPs, reference files, brand rules, research, and prior decisions become scoped context instead of one oversized prompt.

04
Challenge

Build in disagreement

Critical roles review claims, ask for evidence, and escalate deadlocks. Agreement must be earned, not assumed.

05
Verify

Gate the outcome

Your quality rules become explicit checks. The team returns a verdict, risks, and next actions—not just a polished paragraph.

06
Evolve

Learn with your approval

Accepted outputs and confirmed outcomes can propose improvements. You review every change before the team evolves.

Lab product / analyzthis_design

Our first shipped product is a Design Team.

analyzthis_design is working proof from the lab: a design team that does the work humans usually do—ideate, critique, deliberate, verify, and improve inside existing AI tools.

One team. Distinct accountability.

Kavi grounds the work. Noor and Anuj propose competing information architectures. Arjun audits UX and visual quality. Meera checks business impact. Priya sizes feasibility. Zara protects delight. Raj challenges the premise or arbitrates deadlocks.

Output: structured concepts, DesignSpecs, SHIP / REVISE / BLOCK verdicts, ranked fixes, a lite MCP catalog for any IDE, and a human-reviewed evolution loop.

Learn more Inspect the source
KaviCollects project knowledge before opinions form.
Noor ∥ AnujProduce competing minimal and expert concepts.
ArjunChecks task friction, accessibility, hierarchy, and design-system fit.
Meera ∥ PriyaChallenge business value and implementation assumptions.
RajChallenges the premise or arbitrates unresolved structural objections.
DeviLets the host AI voice each specialist without separate API keys.
Lab architecture

More than a folder of prompts.

These are implemented capabilities across Analyzthis lab products. The same architecture powers shipped products and the custom teams we design for clients.

Knowledge

Scoped project memory

Collect source material into a knowledge bank, then give each role only the context relevant to its job.

Planning

Chunked execution

A strong planner stays strong. Chunks pick cheap local, free, or paid models. Slash commands do not — they use the model already in your chat. See how models are picked.

Quality

Adversarial review

Agents cite specific claims, raise grounded objections, and require evidence before accepting prior work.

Control

Hard quality gates

Critical checks never downgrade. A failed standard becomes a blocker and an actionable correction.

Portability

Works where you work

Slash commands in Cursor, Claude, and similar tools — plus a lite MCP catalog for Lovable, v0, Bolt, Replit, and ChatGPT. Custom teams live in the tools your firm already uses.

Learning

Human-reviewed evolution

Accepted work and confirmed outcomes can suggest lessons, routing changes, and reference updates—never silently.

Built for small firms

Real teams for real firms.

We build for MSMEs—architectural studios, CA practices, clinics, and other small firms where the same work repeats every week and partners still review everything.

These are example teams, not off-the-shelf products. Each firm gets the smallest useful roster around the work humans already do—briefs, filings, follow-ups, checklists, and client updates.

Architectural firm

Drawing & submission team

MSME8–20 people

Work humans usually do: turn briefs into drawing sets, track municipal submissions, and prepare site-meeting notes.

  • Check drawing sets against the client brief before partner review
  • Track municipal submission deadlines and missing sheets
  • Compare vendor BOQs before sign-off
  • Prepare site-meeting notes from field photos and mark-ups

Partner still approves every submission.

CA firm

Books & compliance team

MSME5–15 people

Work humans usually do: organize client books, chase missing documents, and prepare filing packets before the partner signs.

  • Organize client books before GST and audit work begins
  • Flag missing invoices and ledger mismatches early
  • Draft client reminder letters for pending documents
  • Prepare monthly partner review packets

Partner still signs every filing.

Interior design studio

Client delivery team

MSME6–12 people

Work humans usually do: build material schedules, compare vendor quotes, and keep site progress aligned with the design intent.

  • Build room-wise material schedules from mood boards
  • Compare vendor quotes line by line
  • Track site progress against the approved design
  • Prepare client presentation decks for approvals

Designer still owns the final look and client relationship.

Dental clinic

Patient care team

MSME4–10 people

Work humans usually do: prepare visit summaries, track treatment plans, and keep insurance and follow-up paperwork moving.

  • Prepare patient visit summaries before the doctor enters
  • Track treatment plans and follow-up dates
  • Organize insurance paperwork for billing staff
  • Send post-visit care instructions consistently

Doctor still makes every clinical decision.

Law chamber

Case preparation team

MSME3–8 people

Work humans usually do: organize case facts, track filing dates, and draft first-pass client updates for senior counsel.

  • Organize case facts from client documents
  • Track filing dates and court deadlines
  • Draft first-pass client update emails
  • Prepare hearing briefs for senior counsel review

Counsel still argues every case.

Manufacturing SME

Orders & accounts team

MSME15–40 people

Work humans usually do: match purchase orders to invoices, follow up on dispatch, and flag stock and payment issues before month-end.

  • Match purchase orders to incoming invoices
  • Track dispatch and delivery follow-ups
  • Prepare weekly stock and reorder notes
  • Flag vendor payment mismatches before accounts closes

Owner still approves payments and orders.

Control and trust

Automation without abdication.

Lab products ship as plain source with consent-based evolution. Custom teams can follow the same principles.

Clear boundaries

Every role has allowed jobs, forbidden jobs, and an explicit output contract.

Human checkpoints

Assess-only modes and approval gates keep implementation under your control.

Model choice

Use host models, local Ollama, free endpoints, or selected paid providers based on the task.

Auditable system

Lab products like analyzthis_design ship readable source with no hidden auto-install behavior.

Questions

Before we map your workflow.

Is Analyzthis just one product?

No. Analyzthis is a lab. We ship focused team products such as analyzthis_design and analyzthis_product, and we also design custom teams when a workflow needs its own roster.

Is this a generic chatbot setup service?

No. The unit of design is the workflow: specialist roles, scoped knowledge, handoffs, failure behavior, quality gates, and outcome feedback. A chat interface may be one entry point, but it is not the architecture.

Does every workflow need many agents?

No. We use the smallest useful team. A narrow process may need two roles and one gate. Extra agents add coordination cost, so every role must earn its place.

Will you replace our current tools?

Usually not. The goal is to work inside or alongside the tools your team already trusts. Discovery includes your current software, data boundaries, and deployment constraints.

What changed in analyzthis_design 2.5.2?

Personas cite real motion and UI libraries and prefer whatever is already in your repo. Kavi records that list on collect. 2.5.1 still stands for anti-slop craftsmanship. 2.5.0 still stands for keep/skip scores. Full list: What's new.

What changed in analyzthis_design 2.4.0?

Claude Code and Desktop spend fewer tokens per verdict. MCP defaults to a small router instead of 19 tools (use mcp --catalog full for the old list). Every MCP turn gets an inferred receipt — not a bill. Skills load a short front; retrieve the rest. Knowledge bank is an index, not a dump. Type /receipt or run npx analyzthis_design receipt. Full list: What's new.

Do I need Cursor or Claude?

No. Cursor, Claude Code, Claude Desktop, Windsurf, Codex and Grok each install with one command. Lovable, v0, Bolt, Replit and ChatGPT install nothing at all — you copy the team in as a system prompt. Pick your tool on the design team page. Beyond that, this applies only to lab products such as analyzthis_design. A custom team for a practice does not ask you to install a coding tool. We start from the software you already open.

Can this run without paid model APIs?

Lab products such as analyzthis_design support a host-AI mode with no separate API keys, plus local and free-model routes. In chat, personas use the model already in Cursor or Claude. Automatic model picking happens only on the CLI run command. Your final setup depends on volume, privacy needs, latency, and required model quality. How models are picked.

How do you keep quality under human control?

We define assess-only stages, approval gates, explicit verdicts, and human-reviewed evolution. The system can recommend a change; you decide whether it becomes canonical.

What happens after I submit the form?

We review the workflow, tools, and desired outcome you describe. If the use case is a fit, the confirmation screen will invite you to book a discovery call.

Start with the workflow

What should your team take off your plate?

Tell us about one repeatable workflow. We will use the detail to prepare a focused discovery conversation—not a generic sales call.

What we collect:Contact details and workflow answers, saved as you go—even if you leave before the last step. Form views are counted on our own server; no cookies or third-party trackers.
Team discoveryStep 1 of 3

First, who are we designing for?