ideally should be
- unixy and composable: https://www.notion.com/blog/data-model-behind-notion
- the chat and repo for inspiration: https://chatgpt.com/share/68650cba-33c0-800d-a518-06fe6e6e8748,
unternet kernel is the middleware that makes the intent router and data model work, but a very good way to hedge on top of mcp and other api systems: https://github.com/unternet-co/client/tree/main/kernel
anthropic subagents architecture design, from amp people
on memory management stack: Mem0: Technical Analysis Report
caveat for tool calls: rag is good for retrieval but not tool calls: mcpn repo + mcp-rag paper brief
security
future quests for ultimate moat in training on reasoning traces
-
Use of primopt + dspy for context and prompt design and programming
- Priompt: managements Priompt (priority + prompt) is a JSX-based prompting library. It uses priorities to decide what to include in the context window. This project achieves a similar goal in separating a templating layer from a logical construction layer written in and compatible with TypeScript-based usage.
- dspy: designign prompts. Provides a great way of automagically optimizing prompts for different models though lacks deterministic control of the prompt important for things like caching and high-throughput, low latency production systems.
- on Evals, a short talk from aie world fair with braintrust as the defacto tool
-
vision for the future on device intent router
