AI Magic - Roadmap
AI is not an afterthought in Neome — it’s the very foundation.
Every layer of Neome, from system design to user interaction, is built to harness AI. Both the meta and data layers are engineered to supply rich semantics and context, making AI intuitive, contextual, and deeply embedded.
In Neome, conversations are the input, output, and programming language. Let’s explore how this turns advanced technology into something invisible — yet magical.
AI Input/Output Pipeline
When a user sends an ad hoc prompt—via WhatsApp, Email, or the Neome interface — it flows through a smart pipeline:
1. Language normalization
If needed, the prompt is translated to English using a language model.
2. RegEx prompt
- The system scans for exact or near-exact matches to predefined enterprise prompts.
- On a high-confidence match, the corresponding action is executed.
- A success message is returned to the user.
3. Prompt to JSON
- All available prompt and action schemas are retrieved.
- The prompt is converted into structured JSON using the schemas and LLM.
- On a confident match, the corresponding action is executed and a response is sent back.
4. Agentic Interception
We’re actively exploring intelligent agents that can dynamically intercept, reason, and orchestrate multistep tasks.
5. Retrieval-augmented generation (RAG)
- Prompts are cleaned and tokenized for search.
- Relevant forms, messages, and comments are retrieved from Neome’s database.
- These are converted into a unified “story” attached to the original prompt.
- The LLM uses this enhanced context to generate a precise and contextual response.
6. Contextual Output Formatting
Final responses are enhanced with deep links, embedding live context back into the reply.
AI Programming
Neome supports NeoScript — a command-line scripting language designed to be both conversationally generated and easily tweaked.
This empowers community developers to build and launch enterprise apps in minutes, no coding background required.
MCP server
Neome is a first-class MCP server and can be invoked from other MCP clients.