Field Notebook
A self-improving research companion for graduate students and knowledge workers that builds a persistent model of your domain, remembers your questions, and gets better at finding and explaining what you need.
Graduate students and knowledge workers doing ongoing research in a specific field
- Persistent domain knowledge graph that builds as you read papers, articles, and notes
- Adaptive search that learns which sources you trust and which questions you ask repeatedly
- Weekly research digest that improves at surfacing what matters to your specific work
- Cross-session memory that lets you pick up where you left off without re-explaining context
Current AI chat tools reset between sessions and don't personalize; the SEED framework's 100 stars on GitHub demonstrates a working technical path to AI that genuinely improves with use.
Elicit alone reports 2M+ researcher users, and memory-enabled AI assistants (Dume, Re:Mind, ChatGPT memory, Claude projects) are an actively trending category — the stateless-chat pain point for ongoing research is validated.Pricing | Elicit: The AI Research Assistant ↗10 Best AI Assistants With Memory in 2026 (Tested) - Dume.ai ↗
Highly crowded space: Elicit, Consensus, Scholise, Atlas, Scite, Research Rabbit, Connected Papers, Perplexity, Scispace, plus ChatGPT/Claude memory features — personal-domain-model is a narrow differentiation that big incumbents can (and are) adding.7 Best AI Research Assistants (2026): Hallucination-Verified ↗AI That Remembers Past Chats: The End of Starting Over Every Conversation ↗
Proven willingness to pay in this segment — Elicit charges $49–$169/user/mo (academic vs industry), but grad students are notoriously price-sensitive and free ChatGPT/Claude are credible substitutes, capping upside for a new entrant.Elicit Pricing 2026: 4 Plans from $0-$169/user/month ↗Elicit Academic vs Industry Costs & Fees 2026 ↗
Persistent, personalized, self-improving AI is a durable trend — every major model is adding memory/projects, and the shift from stateless chatbots to long-term digital teammates is widely reported as the next paradigm.AI That Remembers You: How Persistent Memory Transforms Your AI Experience ↗re-mind | AI That Actually Remembers You ↗
SEED is a real paper/repo but it is an on-policy RL distillation technique for agent training, not a user-facing persistence architecture — the source's claim that it 'demonstrates a working technical path' overstates applicability; building persistent domain models, retrieval, and continual learning is doable but non-trivial engineering.GitHub - jinyangwu/SEED ↗