Do you want to know why Recued did something? Do you want to undo it? The audit record is where you look. Audit is the diary. The server writes it as things happen. It stays on the server. It never goes to the cloud.
#What a run records
Every run writes one diary entry. It says which recipe or tool ran. It says who asked for it and over which channel. It says what was read and what was written. And it says how the run ended. If an outside AI did something, the entry names that AI. Where the action came from travels with the record. So "who did this" always has one answer.
Reactive recipes, the ones that tick by themselves, fire many times a day. Those roll up into one view of the whole process, with the detail of each run inside it. So a busy watcher reads as one line, not a thousand.
#Provenance and the timeline
Audit is a real collection of data. Each entry links to the records it touched. Those links power the timeline. A timeline is one feed, in time order, for one thing: a contact, a deal, or a thread. It shows the raw data, the notes on it, and the run history together. "What happened around this person" is answered from what was recorded. No model has to guess it.
#Memory is a different thing from audit
There are two stores. People mix them up. They are kept apart on purpose:
- Audit is the run record. The server writes it as things happen. It has a size limit. When it grows too big, the oldest entries are thrown out.
- Memory is your knowledge. It holds facts, decisions, likes, and product notes you wrote or brought in. It also holds anything an AI saved with your permission. You own it. Nothing trims it. It is what recall searches when you or an AI asks a question.
Memory lives under Data → Memory. There you can add, edit, forget, import, and export entries.
#Writing an entry that gets found
Recall searches the whole entry. It reads the summary and the body together. It matches on the words the entry contains. If a search finds nothing, recall tries again in a wider way. First it drops filler words. Then it tries any of the words instead of all of them. It also knows word endings. So cancel finds cancelled. But it cannot guess a word you never wrote. 2FA matches nothing in an entry that only says two-factor authentication.
So the best thing you can do is this: end an entry with a keywords line:
How do I enable two-factor authentication?
Security settings, then scan the QR code with an authenticator app.
keywords: 2FA, MFA, OTP, login security, second factorThere is no separate keywords field. You do not need one. The line is just part of the body. Recall indexes it like everything else. This was measured on the entry above. With the line, 2FA, MFA, and second factor login all find it exactly. Without it, they find nothing at all.
Two habits that help:
- Write the summary as the question someone would ask. Do not write it as a file label. "How do I get a refund?" is found by more real searches than "Refund policy".
- Put keywords last. The feed shows the start of each entry. A keywords line at the top makes every entry look like a list of words.
#Two kinds of time
Every entry keeps two times. One is when Recued wrote the fact down. The other, when known, is when the real event happened. For a mail, that is the day it was sent, not the day it was synced. Timelines use real-world order by default. So loading a year of old history does not shuffle your feeds.
Runs are also tagged with a mode: live, manual, or backfill. So a pass over old history does not mess up the list of recent activity.
#Reversibility and control
Big actions are checked before they run. They are written down after. That record is what makes undo mean something. Work that takes a long time shows up in a live activity list. You can stop it from any paired device you allow. If a gated call fails after the decision but before the answer comes back, it is parked as "in doubt". Recued asks you instead of trying again. See Grants and approvals.
#Where to look
In the webclient, Runs is where you see the diary. It shows live runs and past runs. It shows recipes that tick often, rolled up with each tick inside. It shows the line of approvals waiting for you. Open any entry to see its full record: who asked, the channel, the reads, the writes, and how it ended. The history of one thing, like a contact, is on that thing itself, as its timeline.
#Retention and export
Audit history is yours. You choose how long to keep it. You can export the entries. There is no cloud copy. Your server is the only keeper of the diary.
Memory is kept in a different way, on purpose. It has no size limit. It is never trimmed. Knowledge you took the time to write down should not vanish because one busy week filled up the run log. It exports as JSON. It imports back the same way. So you can move a set of notes between your own servers.
#Related
- The warehouse, the data these records attach to.