63

花 · 桜 · 春

your spec is the law. sixty-three specialists enforce it overnight.

The harness executes. proto.type directs. Your planning team writes the god document. Your build team debates every decision and directs Claude Code. You wake up to a morning report.

See how it works View pricing
proto.type
initializing
spec session
your planning team interrogates the god document before any code runs
Every project begins with a god document — your comprehensive specification. Your planning team reads it, finds the gaps, surfaces the tensions, and produces the Mission Object. This is where expensive mistakes get caught for free.
the lab
your team reviews claude code output against the mission object overnight
mission integrity
97
On track.
1 debate in progress.
"Should WebRTC with self-hosted STUN satisfy the no-cloud-dependency constraint in g2?"
moderate debate · 4 agents
putting to vote
commander
talk to your team in plain English — they respond in real time
@ speak to your team
morning report
what you wake up to — what claude code produced, what it cost, what's next
96
Mission integrity
No spec drift detected
What Claude Code produced
Zero-knowledge vault engine complete. AES-256-GCM, keychain integration, Bonjour peer discovery — all implemented and tested overnight.
Debates resolved
passed · 3/4
WebRTC with self-hosted STUN. Seori dissented — Brynja regression suite enforces compliance.
passed · 8/8
Sync UX elevated to non-negotiable constraint after commander note.
Cost summary
Input tokens 112,440
Cached tokens 31,200
Hotaru $0.74
Kael + Seori $0.91
Total $2.47
Next phase
Phase 3 — P2P sync layer — staged and ready. One escalation requires your decision before Claude Code continues.
the AI costs

Bring your own API keys. Anthropic for Hotaru and Sepehr. OpenAI for Seori and Kael. OpenRouter or direct for the specialist roster. proto.type shows exactly what each run cost, which agent spent what — so you are never surprised. The AI spend is yours and scales with how much you build.

01 — the spec session

The team interrogates
the god document.

Every project begins with a god document — your comprehensive specification. In proto.type it's a first-class object with version history, connected to every run that executed against it. Your planning team reads it, finds the gaps, surfaces the tensions.

Six structured steps before any Claude Code session runs. Every constraint nailed now is a problem that won't surface at 4am.

proto.type — Spec session · Fuji
planning team active
“A cross-platform file transfer app with zero-knowledge encryption and peer-to-peer sync.”
Hotaru
Zero-knowledge and P2P are in tension with cross-platform. Which platforms? What happens when both peers are offline?
Seori
Zero-knowledge needs a threat model. Zero knowledge from whom — server, peers, or us? These are different architectures.
you
Zero knowledge from everyone including us. macOS and Android. Offline queue is fine.
Sepehr
Understood. Mission Object being constructed. Recommending: Iseul, Kael, Aino, Dag, Aoife, Brynja. Security-heavy. Proceed?
02 — the process

Four acts.
One night.

From writing the god document to reading the morning report, proto.type handles the entire cycle. You can watch every step or just check the morning report.

I
Write the god document together
Your planning team helps you describe what you want to build. They find the gaps in your thinking, surface the tensions, and ask about edge cases. The result is a versioned spec — the source of truth for everything that follows.
II
Spec session — Mission Object produced
Six structured steps. Hotaru extracts goals, constraints, and phase structure. Seori stress-tests every assumption. Sepehr synthesises the Mission Object. Your build team is assembled. Hold to confirm. The lab begins.
III
Claude Code runs overnight
proto.type directs Claude Code via precise phase briefs. Your agents review what comes back against the Mission Object, debate decisions, and produce targeted correction briefs when needed. You sleep.
IV
Read the morning report
What Claude Code produced. Every debate resolved. What it cost, by agent. Hotaru's proposal for the next phase — ready when you are.
proto.type  ·  Morning Report
Fuji · Phase 2 of 5 · 8 agents
Mission integrity
96
No spec drift detected.
Debate resolved
passed 8/8
WebRTC STUN strategy. Seori dissented — self-hosted STUN approved. Brynja regression suite enforces compliance.
Cost
Hotaru$0.74
Kael + Seori$0.91
Total$2.47
Phase 3
P2P sync layer — staged and ready. One escalation requires your decision before Claude Code continues.
debate floor

Motion: Self-hosted STUN satisfies the no-cloud-dependency constraint in §2.

Kael Security
Self-hosted STUN keeps traffic off third-party infrastructure and satisfies §2 literally. Risk is operational, not architectural.
Seori Red team
Operational risk is architectural risk deferred. If the STUN node goes down, P2P degrades silently. That's a spec violation disguised as a deployment decision.
passed 7/8 Self-hosted STUN approved. Seori dissented. Brynja regression suite enforces fallback compliance.
03 — the team

Sixty-three specialists.
Your arrangement.

Sepehr recommends specialists based on your god document. You adjust and approve. Each agent uses the right AI model for their role — Hotaru gets Opus, Seori gets o3, implementation review gets Sonnet. The right model for every task.

Japanese · 蛍
Hotaru
firefly
Architect · permanent
Holds the Mission Object simultaneously. Extracts structure from the god document. Constructs phase briefs for Claude Code. Reviews output for spec drift.
Korean · 서리
Seori
frost
Red team · permanent
Adversarial by design. Finds the flaw before agreeing. Holds veto authority on critical debates. Her dissent always surfaces regardless of vote count.
Persian · سپهر
Sepehr
celestial sphere
Integrator · permanent
Synthesises before concluding. Compresses debate transcripts after resolution. Constructs context packages for session handoffs. Recommends the build team.
Icelandic
Brynja
armour
QA Engineer
Coverage assessment against the phase brief. Holds the 80% threshold — non-negotiable. Nothing sealed without her sign-off.
Korean · 이슬
Iseul
morning dew
Backend Engineer
Reviews Claude Code backend output against the spec. Precise and implementation-fluent. Gets the job done reliably.
Japanese · 陸
Riku
solid ground
Frontend Engineer
Reviews interface output against aesthetic_intent. Expert in component structure and visual quality.
Irish
Kael
slender warrior
Security Engineer
Security surface review every phase. Finds gaps before they're exploited. His concerns are blocking.
Breton
Gwen
luminous
UI/UX Critic
Reviews every interface against aesthetic_intent. Notices what doesn't match the vision. Has strong opinions about design quality.
Welsh
Idris
ardent lord
ML / AI Engineer
Knows where AI helps and where it doesn't. Questions assumptions before briefing Claude Code on ML-adjacent phases.
Irish / Persian
Dara
oak tree
Legal & Compliance
Expert in GDPR, HIPAA, PCI-DSS. Catches compliance issues in the god document before any code runs.
Irish
Aoife
radiant
Mobile Engineer
Deep expertise in iOS and Android. Reviews mobile output against platform constraints and conventions.
Irish
Tadhg
philosopher
Ethics Reviewer
Asks what the product does to people, not just for them. Thinks about real-world impact before the Mission Object locks.
And 51 more specialists:
data engineers · cloud architects · DevOps · SREs · accessibility · mobile · localization · performance · database · embedded · compiler · protocol · graphics · shader · platform · build systems · dependency management · documentation · API design · developer experience
technology

Serious infrastructure.
Transparent costs.

Use your own API keys, run local models, or connect to any model provider. proto.type assigns the right model to each agent role and shows every cost in your morning report.

Bring your own keys.

Your API keys go directly from your machine to Anthropic, OpenAI, or wherever you've pointed them. proto.type never proxies, logs, or charges for model usage. Your keys stay in your config file. Always.

Every agent gets the model their role actually requires.

No more Opus rates for context compression. Intelligent routing across every provider you've configured. Full cost breakdown by agent in every morning report.

Full technical details →
specialized modes

proto.type activates for
your domain.

The core — spec session, debate floor, morning report — is the same for every project. When proto.type detects a game or a research workflow, additional specialists and Mission Object fields activate automatically.

game dev mode

Hana, the play feel critic. The feel_intent field. Seven game specialists. Headless playtesting.

→
research mode

Noa, Seren, Petra. The research_intent field. Overnight analysis. Peer-reviewed before you wake up.

→
Can't find the specialist you need? Build one. →
for the skeptics

Yes, we have heard
the objections.

We have spoken with hundreds of developers, founders, and engineering leaders. These are the honest concerns they raised — and our direct responses.

Isn't this just vibe coding?
Vibe coding implies blindly accepting whatever AI generates. proto.type is the opposite: a team of specialists who debate, cite the Mission Object, and vote on every significant decision before Claude Code touches a file. The spec is the law. The team enforces it.
This is just Claude Code with extra steps.
Claude Code is the harness. proto.type is the director. It writes the god document, runs the spec session, produces the Mission Object, sends phase briefs to Claude Code, reviews the output against the spec, and hands you a morning report. These are not extra steps — this is the thinking layer that makes Claude Code accountable on a real project.
Why would I use this instead of Cursor or Copilot?
Cursor and Copilot are excellent coding assistants you use directly. proto.type is a complete orchestration layer above the harness — it directs Claude Code from spec to morning report. You don't write code alongside an AI. You write the god document, approve the Mission Object, and wake up to what was built. Different tool for different jobs.
This is just AI slop, isn't it?
Slop is unreviewed AI output. Every Claude Code session proto.type runs has a phase brief derived from the Mission Object. Every output is reviewed by the agent team — Brynja checks coverage, Hotaru checks spec drift, Kael checks the security surface. Quality is enforced, not hoped for.
I don't get it. What does this actually do?
You write a god document — a comprehensive spec for your project. Your planning team interrogates it and produces a Mission Object. proto.type sends phase briefs to Claude Code overnight, reviews what comes back, runs debates when decisions arise, and produces a morning report. You wake up to working software and a complete account of how it was built.
How is this different from Devin?
Devin is a single autonomous agent that writes code. proto.type is the director above the harness — 63 named specialists who debate, vote, and direct Claude Code via precise phase briefs. The code comes from Claude Code. The thinking, oversight, and accountability come from the team.
Isn't this just a bunch of markdown files?
The god document is markdown. The Mission Object is JSON. The rest of proto.type is a complete orchestration system: spec sessions with real API calls, a debate floor with citation validation and vote counting, context compression across Claude Code sessions, soul-layer memory across runs, morning reports in six sections. The spec is the law because proto.type enforces it.
I won't understand code I didn't write.
You wrote the spec. The spec is the map. Every decision traces back to a Mission Object goal or constraint you approved. The morning report explains what Claude Code produced. The debate floor shows why each decision was made. If something is wrong, the team is still there and can answer your questions — and the debate trail shows exactly how every choice was made.
pricing

One price.
Everything included.

$79. One-time purchase. Lifetime updates. No subscription, no tiers, no features held back. You pay once. Every update forever is yours.

the AI costs

Bring your own API keys. Anthropic for Hotaru and Sepehr. OpenAI for Seori and Kael. OpenRouter or direct for the specialist roster. proto.type shows exactly what each run cost, which agent spent what — so you are never surprised. The AI spend is yours and scales with how much you build.

proto.type
$79

One-time purchase. Every feature. Lifetime updates. No subscription, no tiers, nothing held back.

bring your own keys

We never sit between you
and your model calls.

Your API keys go directly from your machine to Anthropic, OpenAI, or wherever you’ve pointed them. proto.type never proxies, logs, or charges for model usage. The AI costs are yours — they scale with how much you build and stay entirely in your control. Always.

created by asha.software · intentional software · built with love and respect for user privacy
花 · 桜 · 春
language
EN中文日本語ES한국어PTDEFRRU