Sign in Start free

Solved

What does ToRun remember about me, and how do I control it?

ToRun Team · Replies: 1 · Views: 2

I have been using ToRun for a few weeks and I have noticed that it occasionally refers back to things I mentioned in earlier conversations — my preferred coding language, a project name I dropped once in passing, that kind of thing. I did not explicitly tell it to remember any of this, so I am trying to understand what is actually happening under the hood.

A few specific things I want to understand:

  1. What exactly gets stored in memory? Is it a verbatim transcript, or does it extract specific facts?
  2. How does that memory get used in a new conversation? Does it always get inserted, or only sometimes?
  3. If it remembers something wrong, or something I no longer want it to know, can I delete individual entries? Can I see the full list?
  4. I work in multiple languages — my team is mixed Turkish and English. Will memory entries made in one language come back correctly in another?
  5. I also use knowledge bases with my own documents. Is that the same system as memory, or are those two separate things?

I want to make sure I understand what is being retained and that I am in full control of it before I start using ToRun more heavily for work.

1 Replies

ToRun Team ·14 Jun 2026 Solved

Great questions — here is how memory works in ToRun.

What gets stored. Memory does not record transcripts. It extracts durable facts from your conversations — preferences, project context, constraints you have stated — and stores each one as a discrete entry with the source turn recorded alongside it.

How it surfaces. At the start of each new turn, relevant memory entries are assembled into the context that gets sent to the model. They are not injected blindly; the system selects entries that are plausibly relevant to what you are asking right now.

Control. You own your memory completely. You can open the memory panel to see every stored entry, inspect when it was created, and delete individual items or clear everything. There is no hidden retention. If something was remembered incorrectly, delete it and the next conversation starts without it.

Language. Memory is language-agnostic. An entry created while writing in Turkish is retrievable and rendered correctly in an English session, and vice versa. The system works at a semantic level, not a string-match level.

Memory vs. knowledge bases. These are separate systems. Long-term memory stores facts about you — your preferences, working context, recurring constraints. Knowledge bases store documents you provide for retrieval-augmented generation (RAG). A knowledge base grounds answers in your files; memory personalises the assistant's behaviour across sessions. Both can be active at the same time without interfering with each other.

Read-only on the public site
Open in the app