v1.1 Intelligence Layer

Intelligence Layer

AutoBuilder gets smarter through private venture memory, approval loops, quality-checked outputs, prompt recipes, and structured startup execution patterns — not a custom LLM trained on your data by default.

Mission Control Research hub Venture Intelligence Graph Trust & privacy

LIVEData: DEMO DATAPRIVATE BY DEFAULTLOCAL FALLBACK READYNO CUSTOM LLM CLAIMOPT-IN TRAINING ONLY

Model Router

One job abstraction — cloud when configured, heuristic fallback always.

LIVEFALLBACK

local_heuristicLIVEShipped in production and reachable on this site.
openrouterLIVEShipped in production and reachable on this site.
openaiREQUIRES INTEGRATIONNeeds a connected account or signed setup before it can run.
anthropicREQUIRES INTEGRATIONNeeds a connected account or signed setup before it can run.
deepseekPLANNEDOn the roadmap; not available in production yet.

Default: openrouter · Fallback: local_heuristic · Max job $0.25

idea_to_venture_genome → deepseek/deepseek-chat (LIVE)
venture_genome_upgrade → deepseek/deepseek-chat (LIVE)
cursor_prompt_generation → heuristic-v1 (LIVE)
claim_safety_review → heuristic-v1 (LIVE)
quality_gate_review → heuristic-v1 (LIVE)
agent_task_generation → heuristic-v1 (DEMO)

Prompt Recipe Vault

Versioned tasks — full prompts stay server-side.

Idea → Venture Genome
LIVEShipped in production and reachable on this site.
Structure raw idea into startup DNA
v1.0 · idea_to_venture_genome · max $0.08
Venture Genome upgrade
LIVEShipped in production and reachable on this site.
Refresh genome after edits
v1.0 · venture_genome_upgrade · max $0.08
Landing page structure
DEMOInteractive simulation or sample output — not a live deployment.
Hero, CTA, trust sections
v1.0 · landing_page_structure · max $0.03
MVP feature plan
DEMOInteractive simulation or sample output — not a live deployment.
Routes, scope, first ship
v1.0 · mvp_feature_plan · max $0.03
Cursor prompt generation
LIVEShipped in production and reachable on this site.
Scoped build/repair packets
v1.0 · cursor_prompt_generation · max $0.03
Build error diagnosis
PLANNEDOn the roadmap; not available in production yet.
pnpm build / TS failures
v1.0 · build_error_diagnosis · max $0.03
Agent task generation
DEMOInteractive simulation or sample output — not a live deployment.
Crew lanes and outputs
v1.0 · agent_task_generation · max $0.03
Support answer
DEMOInteractive simulation or sample output — not a live deployment.
Recovery and FAQ tone
v1.0 · support_answer · max $0.03
Pricing strategy
DEMOInteractive simulation or sample output — not a live deployment.
Tiers and wedge offers
v1.0 · pricing_strategy · max $0.03
Launch plan
DEMOInteractive simulation or sample output — not a live deployment.
Channels and checklist
v1.0 · launch_plan · max $0.03
Investor-safe summary
DEMOInteractive simulation or sample output — not a live deployment.
No hype; scenario language
v1.0 · investor_safe_summary · max $0.03
Quality gate review
LIVEShipped in production and reachable on this site.
Output QA dimensions
v1.0 · quality_gate_review · max $0.03
Claim safety review
LIVEShipped in production and reachable on this site.
Public copy risk scan
v1.0 · claim_safety_review · max $0.03
Venture score review
ESTIMATEDHeuristic or model-generated score — not independently verified.
Viability dimensions
v1.0 · venture_score_review · max $0.03

Model Job Logs

DEMO MODEL JOBS

Task	Provider	Validation	Score	Cost	Label
cursor_prompt_generation	local_heuristic · fallback	valid	71	$0.000	DEMOInteractive simulation or sample output — not a live deployment.
claim_safety_review	local_heuristic · fallback	valid	72	$0.000	DEMOInteractive simulation or sample output — not a live deployment.
idea_to_venture_genome	local_heuristic · fallback	valid	69	$0.000	DEMOInteractive simulation or sample output — not a live deployment.

Input summaries only — raw private prompts are not shown here.

Eval Rubrics

Measurable quality before optimization.

Venture Genome Quality
ESTIMATEDHeuristic or model-generated score — not independently verified.
DNA completeness and founder usefulness
Pass ≥ 72 · 6 criteria
Landing Page Structure Quality
DEMOInteractive simulation or sample output — not a live deployment.
Section order and CTA clarity
Pass ≥ 70 · 6 criteria
MVP Plan Quality
DEMOInteractive simulation or sample output — not a live deployment.
Scope discipline
Pass ≥ 70 · 6 criteria
Cursor Prompt Quality
ESTIMATEDHeuristic or model-generated score — not independently verified.
Files, constraints, acceptance
Pass ≥ 75 · 6 criteria
Build Repair Quality
PLANNEDOn the roadmap; not available in production yet.
Actionable fix steps
Pass ≥ 68 · 6 criteria
Support Answer Quality
DEMOInteractive simulation or sample output — not a live deployment.
Recovery without overclaim
Pass ≥ 70 · 6 criteria
Public Claim Safety
LIVEShipped in production and reachable on this site.
Risky claims flagged
Pass ≥ 80 · 6 criteria
Launch Readiness
DEMOInteractive simulation or sample output — not a live deployment.
Checklist realism
Pass ≥ 68 · 6 criteria
Venture Score Quality
ESTIMATEDHeuristic or model-generated score — not independently verified.
Dimension balance
Pass ≥ 65 · 6 criteria
Investor-Safe Summary Quality
DEMOInteractive simulation or sample output — not a live deployment.
Scenario language only
Pass ≥ 75 · 6 criteria

Quality Scorecards

ESTIMATED unless LIVE EVAL ran.

cursor_prompt_generation
71
Clarity 65
Specificity 66
Buildability 66
Trust 80
Usefulness 66
Next action 80
- · Used local fallback — cloud model not called
DEMOInteractive simulation or sample output — not a live deployment.
claim_safety_review
72
Clarity 65
Specificity 66
Buildability 66
Trust 90
Usefulness 66
Next action 80
- · Used local fallback — cloud model not called
- · Contains guarantee language — review
DEMOInteractive simulation or sample output — not a live deployment.
idea_to_venture_genome
69
Clarity 68
Specificity 62
Buildability 62
Trust 86
Usefulness 62
Next action 76
- · Used local fallback — cloud model not called
DEMOInteractive simulation or sample output — not a live deployment.

Venture Memory Signals

Lessons from approvals, edits, and builds.

Founder approved public copy before trial page publish
Ship public claims only after approval gate passes
approval_granted · approved · DEMOInteractive simulation or sample output — not a live deployment.
Removed “AI employees work 24/7” from marketing hero
Avoid implying autonomous workers unless worker logs prove it
user_edited_output · edited · DEMOInteractive simulation or sample output — not a live deployment.
Venture Genome created from one-sentence idea
Start builds from structured genome, not chat-only
venture_genome_generated · none · LIVEShipped in production and reachable on this site.

Training Data Readiness

Structures only — no training runs in v1.1.

idea_to_venture_genome
PLANNEDOn the roadmap; not available in production yet.
consent: none · private: yes · redaction: pending
Quality 86 · candidate
claim_safety_review
OPT-IN REQUIREDExplicit permission required before this data can be used beyond your request.
consent: anonymous_product_improvement · private: no · redaction: complete
Quality 92 · candidate

Cost Summary

Estimated routing costs — not a bill.

Jobs today

Est. cost today

Fallback jobs

Repair attempts

Failed validation

Budget

Within demo budget envelope

DEMO

AutoBuilder routes cheaper tasks to cheaper models, important tasks to stronger models, and falls back locally when keys are missing or costs should be avoided.

Next Intelligence Actions

Add model provider keys
high
Enable LIVE genome enhancement instead of heuristic fallback only.
Risk: API spend — cap with maxCostPerJobUsd
Persist model jobs to Supabase
medium
Cross-session job history and portfolio analytics.
Risk: Schema migration + RLS review
APPROVAL REQUIREDAutoBuilder prepared this; you must approve before it executes.
Add eval runner for Venture Genome
medium
Replace ESTIMATED scores with LIVE EVAL on golden cases.
Risk: False pass if rubric not calibrated
APPROVAL REQUIREDAutoBuilder prepared this; you must approve before it executes.
Wire approval edits to memory signals
high
Lessons from founder edits improve claim safety and copy.
Risk: Must not log private text in public feeds
Ship consent settings to workspace
low
Founders control opt-in training scope from workspace settings.
Risk: Misconfigured opt-in could confuse users
APPROVAL REQUIREDAutoBuilder prepared this; you must approve before it executes.
Create 100 golden eval cases
low
Benchmark OpenRouter vs OpenAI for idea_to_venture_genome.
Risk: Time investment — do after core router stable
APPROVAL REQUIREDAutoBuilder prepared this; you must approve before it executes.

Cursor Intelligence Packets

Copy into Cursor for precise v1.1 upgrades.

Centralize model calls through runModelJob with provider routing and local fallback.
Cursor packet
Generated brief — select all and copyImplement or extend runModelJob in lib/intelligence-layer/run-model-job.ts. Requirements: taskType routing, local fallback, one JSON repair pass, appendModelJob logging. Do not require API keys for pnpm build. Label FALLBACK vs LIVE honestly.
Select all (⌘A / Ctrl+A) in the box above if clipboard copy is blocked.
pnpm build