> ## Documentation Index
> Fetch the complete documentation index at: https://docs.get-hive.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Learned behaviours: rules, workflows and exceptions
> Execution memory is how Hive remembers the way your team works. Review, activate, pause and edit learned rules, workflows and exceptions, with full version history.
Facts tell Hive what is true about your business. **Learned behaviours** — also called **execution memory** —
tell Hive *how your team works*: the standing rules, the repeatable workflows, and the exceptions where the normal
approach should not apply.
Open **Business Brain → Learned behaviours** (`/brain?tab=rules`). Older links to `/memory` open the same tab.
## Three kinds of entry
Filter the list with the chips at the top:
| Type | What it captures | Example |
| - | - | - |
| **Rules** | Standing instructions, limits and preferences. | "Always copy the account manager on renewal emails." |
| **Workflows** | Repeatable sequences your team follows, including automations you accepted. | "Auto-run: send a reminder when an invoice is 14 days overdue." |
| **Exceptions** | Situations where the normal behaviour must not apply. | "Do not auto-run reminders for this customer." |
Each entry shows **What Hive does**, what it was **Learned from**, how many **Times applied**, and its **Version
history**.
## Where entries come from
Hive learns from what your team does, not from instructions typed into a box:
* **Approvals.** When the same kind of action is approved three times within 30 days, Hive offers an automation.
Accepting it adds an "Auto-run" workflow entry — see
[Autonomy and blast radius](/approvals/autonomy-and-blast-radius).
* **Denials and failed checks.** A **Do not run**, or an action that ran but could not be verified, adds a
shadow exception proposing that Hive stop automating that shape of action. It takes effect once someone activates it.
* **Reflection.** Hive's hourly reflection can propose rules it sees in your data.
Everything Hive learns by itself starts as **Shadow** or **Proposed**. It never activates its own rules.
## Statuses
| Status | Meaning |
| - | - |
| **Active** | Hive applies the entry. |
| **Proposed** | "Hive wants to promote this. Activate to let it run." |
| **Shadow** | "Observing only, proving itself before it acts." Hive records what the entry would have done. |
## Manage an entry
| Action | What it does | Who can do it |
| - | - | - |
| **Activate** | Promotes a proposed or shadow entry to active. Entries from untrusted sources cannot be activated. | Owner, Admin, Department lead |
| **Pause** | Moves an active entry back to shadow. | Operator and above |
| **Edit** the summary | Rewrites what the entry says (1–2,000 characters). Editing an **active** entry drops it back to **Shadow**, so a changed rule proves itself again before it runs. | Operator and above |
Every change adds a row to **Version history**, so the team can see how an instruction evolved and who changed it.
Learned behaviours never override safety. A rule cannot bypass role permissions, blast-radius policy, spend caps
or the kill switch. When the kill switch is engaged, every active entry is demoted to shadow.
## Keep execution memory healthy
* Pause an entry the moment it is wrong; it stays in history and can be reactivated.
* Review entries after a change in pricing, process or responsibilities.
* Prefer one clear entry over several with slightly different wording.
* Check **Times applied** — an entry that is never applied may be scoped too narrowly.
If the list is empty you will see **Nothing taught yet**: "As your team corrects Hive and approves its work, the
rules, workflows and exceptions it learns will appear here — versioned and yours to edit."
## Related
The facts Hive knows.
How approvals turn into earned automation.
Kill switch and policy.