The Problem
Developers lose 66% of AI productivity gains to context management overhead.
Linear Context
Conversations are one-dimensional streams. Context windows fill and truncate. Exploration branches disappear when you try something new.
Amnesia by Design
Every new session is a "brand new hire." Memory features are opaque and platform-locked. No continuity across providers.
No Spatial Navigation
You can't see where you've been or where you could go. Research shows spatial memory improves recall by 28%.
Platform Lock-in
Conversations trapped in silos. Can't move thinking between ChatGPT, Claude, and local models. No portability.
Why Context Windows Aren't the Answer
Models now offer 200K-1M tokens, yet 66% still waste time on context issues. Why?
- Larger context = higher cost per request ($$$)
- "Lost in the middle" problem worsens with length
- No organization = no findability
The real problem is navigation and organization, not capacity.
The Solution
Locus transforms AI chat from linear scroll into visual knowledge graphs with branching, compression, and memory.
locus
Method of Loci
The method of loci is a mnemonic technique from ancient Greece. Orators mentally placed ideas in specific locations within an imagined building—a "memory palace"—then retrieved them by walking through the space.
Locus applies this to AI conversations: place your thoughts in a visual, navigable space. Branch, compress, return. Your memory palace for thinking with AI.
Core Primitives
@checkpoint("hypothesis-v1") # Save conversation state
@branch("security-focus") # Fork exploration path
@switch("hypothesis-v1") # Return to checkpoint instantly
@compress(2) # Free 80% tokens, keep insights
@compare("branch-a", "branch-b") # Side-by-side comparison
@inject("topic:auth-patterns") # Add pre-built context
Key Capabilities
See your entire conversation as a zoomable, navigable DAG on infinite canvas. Galaxy → branch → message.
4-level summarization (full → key → highlights → summary). Always recoverable. Never lose context.
Import ChatGPT, Claude, Gemini. Export anywhere. Your thinking isn't locked to a vendor.
Spaced repetition (FSRS algorithm) surfaces learnings before you forget them.
Market Opportunity
| Metric | Value | Notes |
|---|---|---|
| TAM | $8-10B | AI productivity tools (2025) |
| SAM | $2.4-6B | 40-50M power AI users @ $5-10/mo |
| SOM Year 3 | $1-10M ARR | 15K-100K paying users (conservative) |
Comparable Growth
| Company | Growth | Notes |
|---|---|---|
| Cursor | $0 → $1B ARR in 24 months | Fastest B2B SaaS ever |
| Replit | $2.8M → $253M ARR in <1 year | Post-AI Agent launch |
Target Segments
Developers
Managing context across coding sessions. Highest willingness to pay. Clear pain point.
Researchers
Branching hypotheses, comparing conclusions. Academic and industry R&D.
Knowledge Workers
Building persistent knowledge bases. Writers, analysts, consultants.
Competitive Landscape
Extended analysis including previously missing competitors (February 2026).
| Feature | ChatGPT | Claude | Notion AI | LibreChat | Locus |
|---|---|---|---|---|---|
| Visual Graph | No | No | No | No | Yes |
| Branching | Hidden | No | No | Yes | Yes |
| Cross-Platform | No | Partial | No | Yes | Yes |
| Controllable Compression | No | No | No | No | Yes |
| Cognitive Memory | No | No | Partial | No | Yes |
| Context Window | 400K | 200K-1M | 50-conv history | Varies | Unlimited* |
*Unlimited via hierarchical compression + checkpointing
Extended Competitor Coverage
| Competitor | Pricing | Key Threat | Locus Counter |
|---|---|---|---|
| Obsidian + AI plugins | Free + $4-20/mo | Local-first, graph view for notes | AI-native, not plugin-dependent |
| Perplexity Pro | $20-200/mo | Research focus, citations | Branch hypotheses, compare conclusions |
| Poe | $5-250/mo | Multi-model, 200+ models | Memory that persists, context you control |
| LibreChat | Free (self-host) | Open source, MCP, branching | Visual graph, compression, cognitive memory |
| LobeChat | Free (self-host) | Beautiful UI, plugins | Beyond UI: navigate, branch, compress |
Architecture (Simplified)
Learned from over-engineering critique: PostgreSQL-only until proven insufficient.
Why PostgreSQL-Only
- pgvector handles millions of embeddings with HNSW indexes
- ltree provides efficient tree/hierarchy queries
- Recursive CTEs handle graph traversal for conversation depth
- OpenAI scaled ChatGPT on unsharded PostgreSQL
- Add complexity only when hitting measured bottlenecks
Scaling Path
| Timeline | If Hitting... | Add... |
|---|---|---|
| Month 3 | Connection limits | PgBouncer |
| Month 6 | Read throughput | Read replicas |
| Month 12+ | 10M+ vectors | Dedicated vector DB |
| Month 12+ | Complex graph analytics | Apache AGE or Neo4j |
Moat Analysis (Honest)
Learned from competitive analysis: most "moats" are fake.
Fake Moats (Don't Rely On)
| "Moat" | Why It Fails | Time to Copy |
|---|---|---|
| Better prompts | Easily reverse-engineered, obsoleted by model improvements | 3-6 months |
| Nicer UI | AI generates UIs now; competitors clone fast | 3-6 months |
| "Smarter" context | LangChain/LlamaIndex already free; models improving | 6-12 months |
Real Moats (Build These)
Context Accumulation
More usage → better personalization → harder to switch. User context profiles compound over time.
Team Networks
Team switching cost is 10x individual. Shared context creates network effects.
Open Source Community
Free contributions improve product. Enterprise features justify commercial tier.
Workflow Integration
Deep integration with git, IDEs, CI/CD. Automation chains break on switch.
3-Year Defensibility
| Year | Focus | Moat Built |
|---|---|---|
| Year 1 | User context accumulation + open source | Data flywheel starts; community forms |
| Year 2 | Team networks + workflow integration | Network effects; switching costs increase |
| Year 3 | Vertical specialization + enterprise | Domain expertise + relationships = durable |
8-Week MVP Roadmap
Realistic scope for 2-person team. PostgreSQL-only architecture.
Weeks 1-2: Foundation
PostgreSQL schema + migrations. User auth (JWT). Basic API via PostGraphile.
Weeks 3-4: Core Features
Create/list conversations. Add messages with streaming. Branch creation and switching.
Weeks 5-6: Visual
tldraw integration. Conversation graph rendering. Zoom/pan navigation.
Weeks 7-8: Polish
Import from ChatGPT. Basic compression (levels 0-3). Memory (remember/recall).
NOT in MVP
Business Milestones
| Milestone | Users | Revenue |
|---|---|---|
| Launch | 1K free | $0 |
| Month 6 | 10K free, 300 paid | $2.4K MRR |
| Month 12 | 50K free, 2K paid | $16K MRR |
| Month 24 | 200K free, 10K paid | $80K MRR |