> ## Documentation Index
> Fetch the complete documentation index at: https://slowave.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Procedures: how your agent learns to work

> Slowave doesn't only remember what is true. It also saves how a task was done well, so your agent can reuse proven methods.

Most memory tools store facts. Slowave also stores **procedures**: the ordered steps that worked the last time your agent did a kind of task.

![A task is finished and its working method is saved as a procedure. On a similar task later, the procedure is offered to the agent. The agent reports whether it helped, and the procedure improves.](https://mintcdn.com/slowave-ai/hrOsN_VNHRWDj6ew/images/diagrams/procedures-mobile.svg?fit=max\&auto=format\&n=hrOsN_VNHRWDj6ew\&q=85\&s=269df2b7667ec55043a3d520be873b72)

![A task is finished and its working method is saved as a procedure. On a similar task later, the procedure is offered to the agent. The agent reports whether it helped, and the procedure improves.](https://mintcdn.com/slowave-ai/hrOsN_VNHRWDj6ew/images/diagrams/procedures.svg?fit=max\&auto=format\&n=hrOsN_VNHRWDj6ew\&q=85\&s=6127cd83a45594a45579e3716779804f)

From one finished task to a method your agent can reuse.

## Facts tell your agent what. Procedures tell it how.

> [card] Memories
>
> What is true: decisions, rules, preferences and lessons.
>
> *"Errors always use the `{ code, message }` format."*

> [card] Procedures
>
> How something was done: steps, context and warnings from a task that worked.
>
> *"Add an endpoint: define the route, validate input, write tests, update the API docs."*

## How a procedure is born

**1. Your agent finishes a task**

It closes the task and, if it used a method worth keeping, saves it.

**2. Slowave stores the method**

A procedure has a goal, ordered steps, the situation it applies to, and any warnings.

**3. A similar task comes along**

When your agent starts a related task, Slowave offers the procedure as a head start.

**4. Your agent rates it**

Did it help? Procedures that keep helping are offered more often.

## What a procedure looks like

> [accordion] Example: shipping a hotfix
>
> **Goal:** Ship a fix to production without a full release.
>
> 1. Branch from the last release tag.
> 2. Apply the fix and run the full test suite.
> 3. Open a pull request against the release branch.
> 4. After approval, merge and run the deploy pipeline.
> 5. Watch error rates for 15 minutes.
>
> **Watch out:** Don't merge without at least one reviewer's approval.

## Good to know

> [accordion] Procedures are advice, not orders
>
> Your agent decides what to do. A procedure is a tip from past work. Your agent decides whether to follow it.

> [accordion] Failures are kept as warnings
>
> If a method didn't work, Slowave keeps it and labels it as a caution. Your agent learns what to avoid.

> [accordion] You can see every procedure
>
> The **Procedures** view in the dashboard lists them, with when each was offered and whether it helped.

![Slowave dashboard procedures page listing reusable methods.](https://mintcdn.com/slowave-ai/hrOsN_VNHRWDj6ew/images/dashboard/procedures.jpg?fit=max\&auto=format\&n=hrOsN_VNHRWDj6ew\&q=85\&s=c3e3ab55ee2b28c249dab784d634d734)

The Procedures view in the dashboard. Example data.

> [card] [Next: where your data lives](https://slowave.ai/local-and-private.md)
>
> Everything stays on your computer.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.