---
name: index-gateway-memory
description: >
  Sparse Agent Memory user guide: connect cloud memory, ask a connected Agent
  to save and retrieve working preferences and project lessons, and manage
  projects, query settings, backups and restores in the web workspace.
  Use when helping a user get started or manage their memory.
---

# Sparse Agent Memory user guide

Sparse Agent Memory helps Agents retain working preferences, project conventions and reusable lessons for later tasks. Ask a connected Agent to save new information, or use the web workspace to query, check and back up your memory.

For your first use, save a preference that is easy to verify, then query it and check that it informs later work. The steps below follow that process.

## Choose where to start

| What you want to do | Where to begin |
| --- | --- |
| Let Sparse Agent use memory during tasks | Open **Settings → Cloud memory** in the [Sparse Agent workspace](https://sparse-agent.com/). |
| Query and manage saved memory | Sign in to [Sparse Agent Memory](https://memory.sparse-agent.com/) and open **My projects**. |
| Learn what to remember and when to use it | Read the [cloud memory guide and examples](https://sparse-agent.com/cases/agent-memory.en.html). |

Memory is managed by account and project. The web workspace shows your own projects. In the Sparse workspace, each Agent uses its own memory; connecting to the same service does not automatically share all content among Agents.

## Connect cloud memory in Sparse Agent

1. Sign in to the Sparse Agent workspace and open **Settings → Cloud memory**.
2. Select **Connect Chat account** and authorize with the Auto-Coder Chat account associated with the workspace. You normally do not need to copy an API Key.
3. Return to settings and check the connection status and **Memory query model**. For a first connection, keep **Automatic: keep existing cloud settings first**. If no cloud configuration exists, the Agent model can be reused.
4. Use memory in subsequent tasks for existing or new Agents. The connection takes effect at the next task activation.

After authorizing memory in the Sparse workspace, you can use its cloud memory features without obtaining a separate memory-service invitation. Signing in to the memory website separately still follows that website's joining instructions.

If the page requests authorization again or says model setup is incomplete, resolve that prompt before asking the Agent to save or query memory. You can also choose **Use Agent model** or **Keep separate cloud model**, select **Save model choice** and check the result.

## Save a preference and retrieve it

These examples are for an Agent already connected to memory tools. Begin with a clear instruction:

```text
Please remember: when writing my project reports, start with completed work, then list unfinished items and next steps.
```

In a later task or conversation, ask:

```text
Find my requirements for project reports, then use them to write today's report.
```

Review the retrieved information and the report to confirm that the preference was used. Whether memory is queried or updated depends on the Agent's integration and task instructions. Saying what to remember and when to apply it makes the outcome easier to check.

Good candidates include lasting preferences, confirmed project decisions, reusable methods and recurring mistakes to avoid. Give temporary information an applicable timeframe. Keep passwords and keys in a dedicated credential-management system.

## Query memory on the web

Open [Sparse Agent Memory](https://memory.sparse-agent.com/) and select **Sign in with Auto-Coder Chat**. Invitations are disabled by default: new users register automatically after verifying their Chat identity. If an administrator enables invitations, first-time members select **New here? Join with an invitation**. Existing members continue to sign in directly.

1. In **My projects**, select the project you want to inspect.
2. Open **Query** and enter a question in **Query**, such as “What are my requirements for project reports?”
3. On your first query, keep the default result limit and leave the optional kind blank. Add a kind later if you need to narrow the results.
4. Select **Query** and wait for the result. If the request is queued, follow the status shown on the page.

Check whether the returned content is relevant to the current task. If nothing is found, confirm the selected project and try another phrasing. A newly created empty project has no memory to query yet; a connected Agent must first save content, or you can import an existing backup.

Project details also show information such as the last update. Select **Refresh details** for the latest display or **Back to projects** to choose another project. To manage another area of work separately, enter a new project name in the list and select **Create project**. The Agent using memory must also be directed to the appropriate project.

## Team memory and reviews

Open **Team** to create a team or request membership using an Owner’s join link. The Owner reviews requests under **Members & requests**. Members can read approved memories and propose new entries, updates and deletions.

Review is enabled by default for all submissions, including the Owner’s own. **Memory reviews** compares the current approved version with the proposal. Approval publishes the change; rejection requires a note. Authors can withdraw pending submissions. The Owner can disable review in **Team settings** for future submissions; existing pending proposals still need review.

Select `Team · name` in **Settings → CLI / MCP** to create a credential for that Team. A successful save with a pending state means submitted for review, not published. Searches use the caller’s query settings. Personal projects are not automatically shared with a Team.

Leaving or removal revokes that member’s Team credentials and withdraws their pending submissions. Approved memories remain available to the Team. Owners must transfer ownership before leaving.

## Set up queries when needed

If a query reports missing configuration, open **Settings → Query services**. Keep an existing working configuration. Otherwise, choose a provider in **Model settings**, enter your own API Key and save.

If you already use TypeSafe JEV, you can save your configuration in **JEV settings (TypeSafe)** on the same page. The page shows the effective source. Configured JEV takes priority; otherwise, the saved model is used. If a configuration is invalid, follow the prompt to fix it before querying again.

These settings support memory retrieval. Changing a query configuration does not rewrite saved memory. Export, import, status checks and validation do not require a query model to be configured first.

## Update instructions and migrate existing memory

When circumstances change, tell the connected Agent the new requirement and when it applies. For example: “Send reports every Friday instead. Update the previous schedule and use the new one starting this week.” Query again to check that you retrieve the current requirement.

If a Sparse Agent has local memory, open **Settings → Cloud memory** in the Sparse workspace, choose **Migrate existing memory** and follow progress for that Agent. Migration retains the original local records. It transfers saved memory; chats, browser data and attachments should be saved separately.

To stop cloud access temporarily, select **Disconnect** on the same page. Saved cloud memory is retained. Connect again as prompted when you want to resume using it.

## Check, back up and restore memory

Use these actions in project details as needed:

| Action | What to do |
| --- | --- |
| Status | Open **Status** and select **Load status** to check the project. |
| Validate | Open **Validate** and select **Run validation** to check that the saved content meets the service's requirements. Validation does not modify memory or establish that its content is factually correct. |
| Export | Open **Export**, select the export button and save the complete returned JSON as a backup. |
| Import | Back up the current project first. Open **Import (replace bundle)**, choose an existing JSON backup or paste its complete content, check the target project and select **Confirm replacement**. |

**Import replaces all current memory content in the target project.** Old content absent from the backup will not be retained. For everyday additions or corrections, ask an Agent already connected to memory; use import when you need to restore a backup.

After restoring, check status and query something familiar to verify the result. Backups contain your memory, so save them in a suitable location you control.

## Resolve common problems

| What you see | What to do next |
| --- | --- |
| The page requires an invitation, or a code is unavailable | The administrator has enabled invitation-only registration. Use a valid code from your inviter. Existing members should return to direct sign-in and check the Chat account. |
| Session expired or reconnection requested | Sign in again on the website. For Sparse's memory connection, authorize again in **Settings → Cloud memory**. |
| Query configuration is missing or invalid | Check **Settings → Query services**. When using Sparse, also check the workspace's memory query model. |
| Empty query results | Confirm the account, project and save result, then try different wording. |
| Request is queued | Wait for a status update. After a timeout, read back the result before deciding whether to repeat a write. |
| Import or validation failed | Read the page's explanation and confirm that you selected a complete backup. A rejected import preserves the existing content. |

Once you have completed a first “save → retrieve → apply” cycle, you can use memory in more of your daily work. Continue to the [Sparse Agent guide](https://sparse-agent.com/cases/goal-kpi-constraints.en.html) to set goals, review progress and plan subsequent tasks.

## Claude Code, Codex and WorkBuddy

Use Auto-Coder Gateway CLI, MCP or a Skill with the same memory project. Follow the [CLI / MCP connection guide](https://memory.sparse-agent.com/docs/clients?lang=en) to install, create a project credential and connect your client. Settings → CLI / MCP also provides setup and revocation.
