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

# Use project memory

> Give the Sazabi agent durable context that survives between investigations — browse, write, and organize project memory from the CLI.

Project memory is durable knowledge the Sazabi agent can draw on across investigations — how your system is laid out, what a healthy state looks like, and runbooks for recurring problems. By the end of this guide you will have written a memory document and confirmed it is searchable. It takes a few minutes.

Memory is managed through the `sazabi` CLI (and used by the agent). There is no dashboard page for it.

## What the agent remembers

Project memory is **curated**, not automatic. Sazabi does not silently record everything from your threads; instead, durable knowledge is written deliberately — by you or by the agent when it captures something worth keeping. That keeps memory a high-signal reference rather than a transcript. Raw logs, one-off findings, and hypotheses do not belong in memory.

## Browsing memory

Install and authenticate the CLI first (see [Connect your coding agent](https://docs.sazabi.com/fix/connect-coding-agent.md)), and select your organization and project with `sazabi organizations use` and `sazabi projects use`.

* List the memory catalog:

  ```bash theme={"system"}
  sazabi memory list --json
  ```

* Read a specific document by its path:

  ```bash theme={"system"}
  sazabi memory get notes/investigations/checkout-latency.md
  ```

## Writing memory

Create a document with `sazabi memory create`, giving it a path, a title, and a body:

```bash theme={"system"}
sazabi memory create notes/investigations/checkout-latency.md \
  --title "Checkout latency investigation" \
  --kind note \
  --body "## Summary
Checkout p95 latency spiked after the 2026-08-20 deploy; traced to a slow query.

## Follow-up
Watch the orders service query time after the index change."
```

You can pass `--file <path>` (or `-` for stdin) instead of `--body` for longer documents.

`sazabi memory create` refuses a path that already holds a document. To change an existing document, use `sazabi memory update`, which replaces its body and keeps its kind and title unless you pass `--kind` or `--title`:

```bash theme={"system"}
sazabi memory update notes/investigations/checkout-latency.md \
  --file ./checkout-latency.md
```

`sazabi memory update` refuses a path that holds no document, so neither command overwrites or creates a document by accident.

## Memory kinds

Memory documents have a **kind** that describes their purpose:

* **note** — investigations, findings, and context. This is the default.
* **runbook** — repeatable operational procedures.
* **change** — records of production changes. These are system-managed; the CLI does not write them.

When writing with `sazabi memory create`, use `--kind note` (the default) or `--kind runbook`. Use path prefixes such as `notes/investigations/` and `runbooks/` to keep memory organized.

## Verify

Confirm a document you wrote is searchable.

1. Write a note with a distinctive title, for example:

   ```bash theme={"system"}
   sazabi memory create notes/investigations/memory-verify-8842.md \
     --title "Memory verify 8842" --kind note \
     --body "Verification note for the docs quickstart." --json
   ```

2. Search for it:

   ```bash theme={"system"}
   sazabi memory search "memory verify 8842" --json
   ```

3. The search returns the document you wrote, with its path `notes/investigations/memory-verify-8842.md` and title `Memory verify 8842`.

If the search returns nothing, confirm `sazabi auth whoami` shows the right organization and project, then re-run the write.

## Further reading

> [card] [CLI reference](https://docs.sazabi.com/reference/cli.md)
>
> Every `sazabi memory` command and flag.


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