OpenFounder

Evidence-driven AI co-founder workspace for startup validation.

OpenFounder helps founders move from idea → validated problem → tested solution by treating every claim as an assumption until proven otherwise. It combines structured workflows (phases, evidence ledger, experiments, interviews, risk register) with a multi-provider AI co-founder that refuses to invent sources or customer quotes.

---

Core Philosophy

• Challenge unsupported claims
• Clearly separate: evidence · inference · assumption · hypothesis · recommendation
• Prefer small, falsifiable experiments over large speculative plans
• Never invent market stats, customer quotes, or sources

---

Features

Workspace Modules

• Dashboard – Project health, risk snapshot (open / hot / mitigated), and highest-value next action
• Phases – Guided 10-phase founder journey with required tasks
• Evidence Ledger – Claims with status (unverified → experiment_confirmed), verbatim quotes, sources, and a promotion guardrail: a claim cannot be marked research-backed, customer-confirmed, or experiment-confirmed without the matching evidence.
• Risk Register – Impact × probability scoring, filters, mitigation & next tests
• Research – Structured source logging
• Competitors – Teardowns with strengths/weaknesses/complaints
• Validation / Experiments – Hypothesis → method → success/failure criteria
• Customer Interviews – Mom-Test style logging + verbatim quotes
• Canvas – Visual relationship mapping
• MVP Planner – MoSCoW prioritization
• Roadmap – Dependency-aware planning
• Business Model – Lean Canvas + unit economics
• GTM – Channel mapping & checklist
• Landing Page Studio – Smoke-test copy generation
• Documents – Exportable packages

AI Co-Founder

Multiple providers with configurable models and graceful fallback:

Provider	Notes
Server Gemini	Uses GEMINI_API_KEY (recommended for hosted)
Direct Gemini (BYOK)	Browser → Google; primary + ordered fallback models on 429/503
OpenAI-compatible	OpenAI, Groq, Together, DeepSeek, LocalAI…
Anthropic Claude	Direct browser API
Ollama	Fully local / privacy-first
Offline Rules	100% free, zero network, deterministic heuristics

Settings → AI Providers lets you pick provider, model presets (or custom IDs), Gemini fallback list, temperature, and max tokens. Keys can stay session-only or persist in localStorage.

Keyboard shortcuts:

• ⌘/Ctrl + J → Toggle AI Co-founder panel
• ⌘/Ctrl + K → Command palette

---

Screenshots & Walkthrough

Desktop workspace:

Tablet view:

Mobile view:

Empty state (no project):

Walkthrough

1. Create a project — Open the app, click "Create New Project", and fill in your idea, problem, target market, and founder context. Or click "Load Demo Workspace (TradePulse)" to explore with example data.
2. Work through phases — The sidebar guides you through 10 founder phases. Complete tasks in each phase to advance.
3. Log evidence — In the Evidence Ledger, record claims with status (assumption → unverified → research-backed → customer-confirmed → experiment-confirmed). A claim cannot be promoted without matching evidence.
4. Score risks — In the Risk Register, score each risk by impact × probability. The dashboard highlights "hot" open risks (score ≥ 4) as your next priority.
5. Run experiments — Design falsifiable tests in Validation. Results feed back into evidence confidence.
6. Record interviews — Log customer discovery interviews with verbatim quotes in Customer Interviews.
7. Plan MVP & Roadmap — Scope features in MVP Planner (MoSCoW) and map milestones in Roadmap.
8. Export documents — Generate Markdown or JSON documents from the Documents page, or export your full workspace as JSON from Settings → Data.

---

Tech Stack

• Frontend: React 19 + TypeScript + Vite 6 + Tailwind CSS 4
• Local Database: Dexie (IndexedDB) with Zod schema validation
• State: React Context (ProjectContext + useProject hook)
• Server: Express + optional Gemini (for /api/ai/* routes)
• Testing: Vitest (validation, risk scoring, model defaults, context builder, phases, IDs)
• Deploy: Static build → GitHub Pages (server routes fall back gracefully)

---

Quick Start

git clone https://github.com/Perchemop/OpenFounder.git
cd OpenFounder

bun install          # canonical package manager (bun.lock is committed)

# Full stack (Express + Vite) – recommended
bun run dev

# Frontend only
bun run dev:vite

bun run lint         # tsc --noEmit
bun run test         # vitest run
bun run build:static # production static assets

CI (.github/workflows/ci.yml) runs the type check, the test suite on Ubuntu, macOS, and Windows, and the static build on every pull request. Treat CI as the authority rather than a local run.

Environment

Copy .env.example → .env:

GEMINI_API_KEY=your_key_here   # optional – enables server-side AI

Without a server key the app still works fully offline or with user-provided keys (Settings → AI Providers).

---

How the App Works

1. Boot & local data

1. main.tsx mounts React inside an ErrorBoundary.
2. ProjectProvider loads settings from localStorage, opens Dexie (OpenFounderDB), and optionally activates an existing project (never auto-seeds synthetic data without consent).
3. Entity tables (evidence, risks, experiments, …) load for the active project ID.

2. Layout shell

App.tsx is a thin shell:

• Navbar – project switcher, theme, settings, AI toggle
• Sidebar – workspace tabs with badges (unverified evidence, active experiments, open risks)
• Main – active page component
• AICofounderPanel – slide-over chat
• Modals: New Project, Settings, Command Palette

3. Validation workflow (intended path)

1. New Project – capture problem, user, constraints, starting point → stored in IndexedDB.
2. Phases – work through 10 founder phases; complete tasks and advance phase index.
3. Evidence – log claims; status starts as assumption/unverified; promote only with real signals.
4. Risks – score impact × probability (1–9); track mitigation and next tests. Dashboard prioritizes “hot” open risks (score ≥ 4).
5. Research / Competitors / Interviews – structured inputs that feed the AI context string.
6. Experiments – falsifiable tests with success/failure criteria; results feed evidence.
7. MVP / Roadmap / GTM / Business / Landing – plan only after riskiest assumptions are under test.

4. AI co-founder

1. buildProjectContext() serializes workspace data (within token budget).
2. User message + context + strong system prompt go to the active AIProvider.
3. Providers: Server/Direct Gemini (with model candidate fallback), OpenAI-compatible, Anthropic, Ollama, or Offline Rules.
4. Offline mode returns deterministic guidance from local counts (no network).

5. Privacy & backup

• All project data lives in browser IndexedDB.
• Export/import full JSON packages from Settings → Data.
• API keys: session-only by default, or opt-in localStorage.

6. Hosting modes

Mode	Command	AI
Full stack	npm run dev	Server Gemini if GEMINI_API_KEY set; else BYOK / offline
Static (GitHub Pages)	npm run build:static	Direct Gemini / OpenAI / Anthropic / Ollama / Offline only

---

Architecture Highlights

State (src/context/ProjectContext.tsx)

Centralized project, entities, AI config, theme, modals. Derived: evidence counts, open risks, AI context string.

Data (src/db/)

Dexie tables, Zod validation, export/import with optional ID remapping, settings persistence.

AI (src/ai/)

• providerTypes – interface + system prompt
• providersGemini / OpenAI / Anthropic / Ollama / Offline
• modelDefaults – presets, buildGeminiModelCandidates, parse/format fallbacks
• contextBuilder – workspace → prompt text

Risks (src/utils/riskScoring.ts + RisksPage + Dashboard snapshot)

Pure scoring helpers; full CRUD register; dashboard strip for open/hot risks.

---

Tests

npm test

Coverage includes:

• Zod / export validation
• Workspace referential integrity (canvas edges, experiment → evidence, roadmap dependencies)
• Risk scoring (riskScore, hot/open, sort)
• Evidence integrity promotion rules (evaluateEvidencePromotion, integrity summary)
• Model defaults (buildGeminiModelCandidates, parse/format fallbacks)
• buildProjectContext
• FOUNDER_PHASES structure
• generateId uniqueness & format

---

Project documentation

Document	Purpose
docs/project-brief.md	Product identity, user, problem, success definition
docs/scope.md	In-scope and out-of-scope capabilities
docs/inspiration.md	Public frameworks and clean-room boundary
docs/architecture.md	Stack, runtime shape, data model, domain rules
docs/capability-inventory.md	The capability list this app is designed to deliver, with sources and status
docs/parity.md	Capability matrix with honest status
docs/releases/	Per-release legal checklist records with CI evidence
docs/assumptions-and-questions.md	Assumptions and open questions
docs/development.md	Contribution flow and CI-only verification
docs/qc/manual-qc.md	Manual quality checklist
IMPLEMENTATION_PLAN.md	Vertical-slice plan
REQUIREMENTS_TRACEABILITY.md	Requirement → code → test mapping
DECISIONS.md	Decision log
PROGRESS.md	Session log with verified results
CHANGELOG.md	Release notes
ATTRIBUTION.md	Third-party licences and asset provenance
CONTRIBUTING.md	How to contribute, and the licence terms for PRs
ROADMAP.md	Readiness percentages and staged milestones

---

Roadmap

• ProjectContext extraction
• Error Boundary
• Risk Register UI + Dashboard snapshot
• Expanded unit tests
• Configurable model lists, temperature, max tokens
• Evidence integrity guardrails + claim editing
• Pull-request CI on Ubuntu, macOS, and Windows
• Documentation and tracking baseline
• Dashboard evidence-integrity score
• Full-text search across research, interviews, risks, roadmap
• Accessibility audit (human pass recorded, automated axe scan in CI)
• Installable PWA with an offline app shell
• Optional cloud sync (still local-first)

Contributing

Contributions are welcome. Please see CONTRIBUTING.md for how to file bugs, propose features, set up a dev environment, and submit pull requests. The project aims to respond to non-security issues within 7 days and prioritises accessibility reports. Review is public; code-of-conduct and security issues are handled privately.

License

Dual licensed under MIT OR Apache-2.0, at your option — see LICENSE, LICENSE-MIT, LICENSE-APACHE, and ATTRIBUTION.md. Contributions are accepted under the same terms (CONTRIBUTING.md).

---

Built for founders who prefer evidence over optimism.