View and manage a playbook

Manage a playbook's FAQs, skills, versions, proposals, and settings in Redis Cloud, and test what agents retrieve.

After you create a playbook, select Agent Playbooks under Iris in the Redis Cloud console menu to see the list of your playbooks.

The list shows each playbook's Status, Name, Playbook ID, when its configuration was last updated, and its database. Select a playbook to open its Configuration tab.

A playbook's page has the following tabs:

• Configuration: the playbook's settings, credentials, extraction, and retrieval policy.
• Registry: the playbook's frequently asked questions (FAQs) and skills, and their versions.
• Proposals: suggested FAQs and skills waiting for review.
• Playground: a test of what an agent retrieves for a query.

If your role can't read the playbook's content, you see only the Configuration tab.

To connect an agent, select Connect on the playbook's page. See Give an agent access.

Registry tab

The Registry tab lists the playbook's entries, with FAQs and Skills on separate tabs. For each entry, it shows the question or skill name, which version is published, the latest version, and when it was updated. Filter the list by Publication to find entries that are published, not published, or have a newer version waiting.

Add a FAQ

1. On the FAQs tab, select New FAQ.

2. In the New FAQ dialog, enter the Question and the Answer your agents should give.

3. Optionally, open Optional settings and set the following:

   Setting	Description
   Alias (optional, immutable)	A stable, human-readable identifier for the entry, using lowercase letters, numbers, and single hyphens, up to 64 characters. You can't change it after you create the entry.
   Retrieval intents	Up to five task paraphrases, one per line, returned with the entry to help an agent interpret it.
   Filter attributes	A JavaScript Object Notation (JSON) object of up to 32 attributes, such as locale or audience, that retrieval filters can match.
   Metadata	JSON for any other context your agent or application needs. Retrieval filters don't search it.

4. Select Create FAQ.

The console saves the FAQ as an unpublished version 1. Agents don't receive it until you publish it.

Upload a skill

A skill packages instructions and supporting files for a procedure, in the Agent Skills format.

1. On the Skills tab, select Upload Skill.
2. In the Upload Skill dialog, drop or browse to an Agent Skills ZIP file, up to 5 MiB. The ZIP must contain one folder, named after the skill, with a SKILL.md file that begins with YAML name and description fields, plus any resources. Exclude __MACOSX files.
3. Optionally, set the entry's Optional settings, as for a FAQ.
4. Select Upload Skill.

See Manage skills for how to package a skill.

Publish a version

Every change to an entry creates a new version. Agents receive only the version you publish, so you can prepare and review a new version while agents keep using the current one.

Select an entry to open its page. The published version card shows which version agents receive, and whether a newer version isn't published yet. The Versions table lists every version and its state.

• To publish a version, select it and select Publish version, or select Publish in the Versions table. The Versions table also has a Withdraw button for the published version.
• To roll back, publish an earlier version.
• To stop serving the entry, select Withdraw version. Its versions stay available, and you can publish one again later.
• To edit the entry, select New version. Saving creates an unpublished version.
• To see what changed between two versions, select Compare versions.

You can also publish an entry's newest version from the Registry list: select Publish latest from the entry's actions menu. The option shows the newest version number, such as Publish latest v3.

Delete an entry

Deleting an entry permanently removes all its versions and resources.

1. In the Registry list, select Delete entry from the entry's actions menu. To delete several entries, select them and select Delete selected.
2. Select the confirmation checkbox, and then select Delete.

You can also delete an entry from its page, in the Danger zone.

Proposals tab

The Proposals tab lists FAQ and skill proposals from agents with propose access. Automatic FAQ extraction adds FAQ proposals only.

The Inbox lists open proposals, with how much evidence supports each one and when to review it by. Select a proposal to review its candidate and Supporting evidence.

For a FAQ, review the question and answer. For a skill, review its name, description, and Instructions. Use Download under Resources to inspect its supporting files. The console verifies that downloaded files match the reviewed proposal.

After review:

• Select Accept as draft FAQ or Accept as draft Skill to create an unpublished version from the candidate. Review it, then publish it when it's ready.
• For a FAQ, select Edit candidate to change the question, answer, alias, intents, or attributes before you accept it.
• Select Move to trash to dismiss it.

If the proposal changes during review, select Reload proposal and review the updated candidate and files before accepting it.

More evidence means more observations, not greater accuracy. Check the candidate against its sources before you accept and publish it. For review through the API, see Review proposals.

The Trash lists dismissed proposals until they're purged. Select Restore to inbox to review one again, or Purge proposal to delete it permanently.

Playground tab

The Playground tab shows what an agent would retrieve for a query, using only published versions.

1. Enter a Query, such as a question your agent needs to answer.
2. Optionally, open Options to choose a Search mode (combined discovery, FAQ search, or skill search), a Result limit, pinned entries, or an Attribute filter. These options apply only to this test. They don't change the playbook's policy.
3. Select Run search.

What your agent sees lists the matching FAQs and skill previews, with each one's version and similarity score, and the full response as JSON. To change how discovery selects entries for every request, edit the discovery policy.

Configuration tab

The Configuration tab has the following sections. To change a section, select its Edit button, make your changes, and select Save changes.

General settings

Shows the Playbook name, which you can change, and the Playbook ID, Embedding model, and Redis Cloud database, which you can't. It also lists the playbook's Agent endpoint for each region.

Embedding provider credential

Appears only for a playbook that uses a Bring Your Own embedding model. It shows whether a provider API key is configured, and lets you replace it.

Automatic FAQ extraction

Extraction turns conversations in Agent Memory into FAQ proposals for review. The playbook must be Ready, and you need access to manage the source service. Each playbook reads one whole Agent Memory service in the same account; you can't select individual namespaces.

1. Enter the Source RAM store ID, the ID of the Agent Memory service whose conversations to read, or choose one from your account. You can't change it after you create the extraction job.
2. Set the Session inactivity window, how long after a session's last event it becomes eligible for extraction, from 1 to 1440 minutes. The default is 30 minutes.
3. Select Create extraction job.
4. Turn extraction on when you're ready.

Extraction suggests FAQs only. It never publishes them: accepted proposals become unpublished FAQs that you review and publish. To stop extraction and remove its settings, use Remove extraction job in the Danger zone.

Turning extraction off pauses new work and keeps the source connection and settings. Work already in progress can finish. A Ready extraction configuration does not confirm that a worker has processed a conversation.

Keep source conversations available until extraction and review finish. A conversation becomes eligible after the inactivity window; that window is not a completion deadline. Active conversations, conversations without reusable answers, and answers already covered by published FAQs might produce no proposal. Extraction can also add evidence to an existing open proposal. Later events in a completed conversation do not automatically trigger extraction again.

Discovery policy

Controls how discovery selects entries for every request.

See Configure discovery for how these settings work together.

Proposal review policy

Set the Review window, how long proposals stay available for review, from 1 to 365 days. Maximum open proposals shows the inbox's capacity, which you can't change in the console.

Danger zone

You can clear entries or proposals only while the playbook is Ready. To confirm clearing or deleting, select the confirmation checkbox in the dialog. None of these actions can be undone. If a clear fails partway, check the playbook's current state before you retry.

Troubleshoot a playbook

Check the playbook's state in Configuration before retrying a failed operation. Keep its identifier and safe error details for support; do not include API keys or conversation content.