One question I often ask myself while building Renma is:
“How is this different from everything else?”
There are already many technologies around LLMs.
Prompt Engineering.
RAG.
Vector databases.
Agent Memory.
Knowledge bases.
So where does a Context Repository fit?
I don’t think these technologies compete with each other.
I think they solve different problems.
Prompt Engineering
Prompt Engineering is about communicating with a model.
How should I ask?
How should I structure the instructions?
How can I get a better answer?
Good prompts are still valuable.
But prompts usually represent one interaction.
They are not where long-term knowledge lives.
RAG
RAG is one of the most practical technologies in today’s LLM ecosystem.
Instead of relying only on the model’s internal knowledge, it allows applications to retrieve relevant information from external sources.
Without RAG, many enterprise AI applications would simply not work.
However, RAG answers a different question.
How do we find relevant information?
A Context Repository focuses on another question.
How should reusable knowledge be organized before it is retrieved?
I think these two ideas naturally complement each other.
A well-organized Context Repository should make retrieval more reliable, not replace it.
Vector Databases
Vector databases are excellent at finding similar information.
They make retrieval scalable.
However, similarity is not the same as structure.
Knowing that two documents are similar does not tell us:
- which one is the source of truth
- which one is outdated
- which documents should evolve together
- why they are related
These relationships are difficult to represent using embeddings alone.
Agent Memory
This is probably the closest concept.
Agent Memory allows AI systems to remember previous interactions, observations, or learned experiences.
I think this is becoming an important part of AI systems.
But the focus is different.
Agent Memory grows through the agent’s experience.
A Context Repository grows through human knowledge.
One captures experience.
The other captures intention.
I believe both are necessary.
Knowledge Bases
Most companies already have knowledge bases.
They contain valuable information.
The challenge is that they are usually written for humans.
A Context Repository is different.
Its purpose is not just storing knowledge.
It is organizing knowledge into reusable context assets that can be reviewed, combined, maintained, and consumed effectively by LLMs.
How I currently see the ecosystem
This is the mental model I have today.
- People create and review knowledge.
- A Context Repository organizes that knowledge into reusable assets.
- RAG retrieves the right assets when they are needed.
- LLMs and agents consume the retrieved context.
- Agent Memory records new experiences from those interactions.
- Valuable experiences may eventually improve the human knowledge again.
Every layer has a different responsibility.
That’s why I don’t see them as alternatives.
I see them as complementary parts of the same ecosystem.
This is only my current understanding
The AI ecosystem is evolving incredibly fast.
Maybe this picture will look different in two years.
Maybe someone will come up with a much better approach.
I’m completely open to that.
At least today, though, this mental model helps me think about where Renma belongs.
Not as a replacement for Prompt Engineering.
Not as a replacement for RAG.
Not as a replacement for Agent Memory.
Just another layer that I believe deserves more attention.
Models will continue to evolve.
Tools will continue to evolve.
I believe well-organized knowledge should continue to provide value.
People │ ▼Knowledge │ ▼Context Repository │ ▼RAG / Retrieval │ ▼LLM / Agent │ ▼Agent Memory │ ▼Experience │ └───────────────┐ ▼ Improve Knowledge
~ my tech diary ~ 