i80agent is a platform in progress for turning private knowledge into grounded, domain-specific AI agents — agents that can answer accurately, reason within context, and grow as your knowledge grows.
Not fully self-serve yet. Today, new agents still require some manual setup. The goal is to make knowledge-based agent creation accessible to anyone.
General LLMs know the internet. They do not know your business, your process, your stories, or your domain unless that knowledge is collected, structured, and retrieved.
i80agent is a platform for building knowledge-based AI agents from private domain knowledge. Users start with a focused knowledge base, keep adding knowledge over time, and create agents that answer from trusted content instead of guessing.
Cooking stories, memories, notes, decisions, creative process, and lived experience.
Hotel guest support, HR onboarding, internal help desks, product documentation, and service FAQs.
LLMs are powerful, but without your private knowledge, they are incomplete. i80agent is designed to ground answers in your own trusted content.
A hotel, restaurant, product, team, or personal archive each has its own rules, vocabulary, context, and answers. The agent should know that world deeply.
A knowledge agent does not need a perfect knowledge base on day one. It can start small, learn from real questions, and improve continuously.
The agent combines structured knowledge, semantic retrieval, orchestration, and controlled use of LLMs.
Collect facts, stories, documents, FAQs, policies, or process knowledge into a structured foundation.
Use embeddings and semantic search to find the most relevant knowledge for each question.
Decide whether to answer directly, ask for clarification, call the LLM, or route to a workflow.
Generate clear answers grounded in trusted content, with boundaries to reduce hallucination.
The system is working — but still evolving.
What started as a personal experiment has grown into a broader exploration of how domain-specific AI agents can be built, refined, and eventually made accessible to others. I am building this in my spare time, so progress comes in waves and moves more slowly than I would like.
Built the first working agent based on my personal cooking stories.
Extended the same architecture to a real-world use case:
Rebuilt the Alex Cooking agent into the Spontaneous Cooking by Alex Set Menu dinner experience, running it locally to guide and enrich dinners with guests.
Continued improving the hotel knowledge base with customer feedback and real usage patterns.
As AI evolved rapidly, I paused active development to rethink the foundation:
This led to a new direction: rebuilding i80agent using tools like Claude Code and OpenAI Codex, with the goal of minimizing manual coding and focusing on system design instead.
Began rebuilding the i80agent platform with ChatGPT, OpenAI Codex, Claude Chat, and Claude Code, using AI-assisted development to move faster from system design to working code.
Rebuilt the i80.com website using OpenAI Codex without writing a single line of code manually.
The goal is to evolve i80agent into a generic platform for building domain-specific AI agents.
Current focus areas:
It started with a simple realization: LLMs are powerful, but they do not know your personal or business knowledge.
That led me from cooking stories and memory into a broader idea: AI agents grounded in real, trusted domain knowledge.
Read My Road to i80A living journal of experiments, product thinking, technical notes, and research on trusted domain agents.
A practical reflection on why the concept behind RAG still matters, even as naive RAG becomes insufficient for domain-specific AI agents.
Why fast AI coding breaks as systems grow, and how structured intent helps AI-generated software scale.
A reflection on the shift from writing code to designing intent through structured natural language.
A comparison of chunk-based and query-focused embedding strategies for private-domain agents.
Research notes on how embedding construction affects semantic search precision.
A Comparison of Query-Focused Embeddings vs. Traditional Embedding Strategies for Domain-Specific AI Agents
A setup note from the first phase of the i80agent journey, when I was getting the Python foundation in place.
i80agent is still in progress. If you are working on similar problems, have a private-domain use case, or want early access, I welcome conversation.
Email Alex