How Lorah
remembers.
Lorah memory is not chat history. Chat history is a record of what was said. Lorah memory is the set of facts, decisions, terms, context, and working assumptions that you choose to keep as durable project knowledge.
Most AI tools are useful inside one session. Lorah is designed for work that continues across sessions, tools, models, teammates, documents, and decisions.
The problem · AI work keeps resetting
When you use several AI tools, the work often gets scattered.
- One model helped with strategy.
- Another reviewed a document.
- A coding assistant made an implementation assumption.
- A research tool found useful evidence.
- A meeting produced a decision.
- A later session no longer knows any of that unless you explain it again.
That is the context tax.
Lorah’s memory system is built to reduce that tax without turning every conversation into permanent truth.
Lorah memory is reviewed, not automatic
Lorah does not treat every message as memory. Memory is created through a review process.
A teammate may notice something worth remembering, such as:
- a project fact
- a decision
- a definition or glossary term
- an important constraint
- a rejected option
- a source-backed conclusion
- a useful working assumption
When that happens, Lorah proposes it. The proposed memory goes to the Review Queue. You decide whether to approve it, edit it, reject it, or leave it out.
Only approved memory becomes durable project memory. This keeps the user in control. Lorah can suggest what may matter, but it does not silently rewrite the shared truth of the workspace.
What Lorah remembers
Lorah memory is designed around serious project work. It can hold:
Facts
Stable information about the project, product, customer, market, system, or situation.
The beta launch is focused on founders, consultants, operators, researchers, and AI power users.
Decisions
Choices that have been made and should not be reopened without a reason.
Lorah should be positioned as a local-first AI workspace, not as an autonomous agent platform.
Glossary terms
Project-specific language that teammates need to understand consistently.
Context tax means the repeated work of re-explaining the same project background to different AI tools.
Constraints
Rules or boundaries that shape future work.
Workspace data should remain local-first. Lorah should not host customer workspace memory.
Working context
Useful background that helps a teammate answer with the right frame.
The user wants launch copy to be practical and precise, not hype-driven.
Every memory has provenance
Approved memory should not be a mysterious note floating in the system. Lorah keeps memory connected to where it came from — a conversation, sync, meeting, council, document, or source.
That means you can inspect why Lorah believes something, not just see that it believes it. This is important for trust. If a memory is wrong, stale, too broad, or no longer useful, you can correct it.
Memory can change
Projects evolve. A decision from last month may be replaced. A launch date may move. A competitor’s pricing may change. A technical assumption may become false.
Lorah memory is built around that reality. Memory can be:
- active
- rejected
- superseded
- disputed
- archived
- removed
The goal is not to remember everything forever. The goal is to keep the current working truth visible, reviewable, and useful.
Memory is local-first
Lorah is a desktop workspace. Your workspace memory lives on your machine.
Lorah’s servers do not host your project memory, conversations, documents, teammates, or library.
Model calls still go to the model providers you choose — OpenAI, Anthropic, Gemini, Perplexity. Lorah does not relay those calls through its own inference service and does not add markup to model usage. You bring your own API keys. Your workspace stays local.
Memory helps teammates work together
Lorah teammates are persistent specialists. A strategist, builder, researcher, critic, and writer should not all need to be briefed from scratch every time.
When memory is approved, Lorah can use it to give the right teammate the right context at the right time. That helps a teammate understand:
- what has already been decided
- what assumptions are current
- what terms mean inside this project
- what constraints should not be ignored
- what evidence supports prior conclusions
- what should not be repeated
This is how work starts to compound.
Memory is not magic
Lorah memory does not mean the system is always right. It does not remove the need for judgment. It does not make every output trustworthy by default.
It gives you a structured way to preserve, inspect, correct, and reuse the context that matters. The user remains the authority. Lorah’s job is to reduce the cost of carrying context across tools, models, and sessions while keeping control visible.
The short version
Lorah memory works in five steps:
- Work happens across teammates, documents, meetings, councils, and tools.
- Lorah identifies information that may be worth remembering.
- Proposed memory goes to the Review Queue.
- You approve, edit, or reject it.
- Approved memory becomes source-linked project context that future teammates can use.
That is the core idea: your work should not reset every time you open a new AI session. Lorah helps your thinking continue.