How a bot remembers
A bot's journal, memory, brief and skills, and how it rereads its own past instead of forgetting what it did.
A bot remembers in layers, like a colleague: who it is and how you like things done, what it is working on, and a journal of everything it already did. Every layer works the same for a bot on this computer and on Vato Cloud.
| Layer | What it holds | Who writes it | Where to see it |
|---|---|---|---|
| Profile | Role, instructions (personality, rules) | You, or the bot when you ask for a lasting change | Actions → Edit profile |
| Memory | Lasting facts and preferences you confirmed | You, or the bot when you confirm a fact | Actions → Memory |
| Brief | What the bot knows about its mission | The bot, during its first questions and when you correct it | Actions → Memory |
| Skills | Procedures it learned, step by step | The bot, after a workflow worth repeating | Actions → Skills (Bot skills) |
| Journal | Everything it did, turn by turn | Vato, automatically | The bot reads it; see below |
The journal
After every turn of every conversation, background work, routine run and delegated task, Vato writes an entry in the bot's journal. Nothing depends on the bot remembering to take notes. Each entry keeps:
- what was asked;
- what the bot did: the tools it used with their targets, for example Edit ×2 (src/App.tsx, …), browser_navigate (https://…), Bash (npm test), and how many tool calls failed;
- the result: the beginning and the end of its final answer;
- the files it wrote or sent;
- whether the turn finished, failed, was cancelled or was interrupted.
Before each reply, the bot gets the latest journal entries from all its sessions, plus older entries related to your message. So in the conversation it knows what a background work found, and in a work it knows what you discussed in the conversation.
The bot can reread its past
When you mention something earlier ("like last time", "where are we on the audit?"), the bot is told to check its journal and search before answering, never to claim it doesn't remember:
- it searches everything it did and said, across the conversation, works, routines and tasks;
- it lists its sessions and rereads any of them, message by message, with a summary of the tools used before each reply.
A bot only reads its own sessions, never another bot's or your agent conversations. A bot in a bot group only rereads that group: what you told it privately never shows up there.
Picking up where you left off
- Your conversation keeps the bot's provider session, like an agent conversation: the provider keeps the whole exchange, tools included, and condenses it when it gets long.
- If that session can't be resumed (another account was used, or the provider lost it), the bot doesn't start from a blank page: Vato gives it the recent conversation, its journal and an automatic summary of older exchanges.
- On Vato Cloud, the conversation keeps its session even after the machine sleeps and wakes up.
- Long answers are kept with their beginning and their end, so a conclusion is never cut off.
Long conversations
A provider condenses a conversation that gets too long: older details can be lost. Vato makes sure the bot keeps its goals, on this computer and on Vato Cloud:
- Automatic summary: every few exchanges, Vato updates a short summary of the conversation, its works, routines and tasks (your goals, preferences, decisions, open questions and next steps).
- After a condensation, the next turn gives the bot back that summary and the journal entries of the condensed part, so what it did before stays in view.
- Open works are listed at every turn of the conversation while they run, wait for your answer or haven't posted their report, so the bot follows up on them instead of starting the same job twice.
- Goals and decisions go in the brief: when you set or change a goal, a constraint or a lasting decision, the bot is told to write it in its brief right away (Actions → Memory), where it survives any condensation.
What the bot forgets on purpose
- When you restore a message or a version of the conversation, or continue a work from an earlier message, the journal entries of the undone turns are removed: the bot doesn't believe it did work you just undid.
- Deleting a work deletes its journal entries; deleting a bot deletes its whole journal.
- The journal keeps the 5,000 latest entries of each bot; older ones can still be found in the conversations themselves.
Turn memory off
In Actions → Integrations, turning Bot memory off means the bot keeps no journal, can't search its past and can't change its memory or brief. Its conversation still resumes its provider session.