There is a lot of similarity, especially regarding the capabilities, between Zive Assistants and Zive Agents.
But there is one major difference: While Zive Assistants are always manually triggered by a user prompt, Zive Agents run fully autonomously once configured. This way you can automate recurring tasks and invest your own time more efficiently and meaningfully.
How to create a Zive Agent
Agents have their own section which you find in your navigation bar. Here you always have a clear overview separated between agents that you activated for yourself and other available agents that you personally are not using yet.
In the top right corner you find the button "Create new" to create a new agent.
After clicking the button you can choose in which space you want to create the agent. This defines who can access your agent after you publish it.
Your space (Personal):
Every user has the permission to create an agent in their personal space
Agents created in your personal space are only available to you and no one else can activate them for themselves.
Company space:
Only users with the platform role "Moderator" can create agents in the company space
Agents created in the company space are available for activation to all users
Team spaces (if existent):
Only users assigned as "Managers" of the respective team space can create agents in the space
Agents created in a team space are available to all members of the respective space.
Tip: You can also navigate into a space and create an agent from there. To learn more about Spaces and their configuration, read our dedicated help article.
In the slider that opens you can give basic information about your new agent. You need to provide a name and a description and can select an icon and the color for your agent. You also have the possibility to define Managers that should be able to edit the agent. All of the information can be adjusted afterwards as well. Click on "Continue" to start defining the detailed settings for your new agent.
How to configure an agent
Tip: Use the preview on the right side to test how your adjustments to the configurations impact the behavior and results. Simply click the button "Execute now" to trigger a run at any time.
Settings
In the first tab you can adjust the base information you just defined, like name and description. Those settings are quite self-explanatory, but in addition to that you can configure the access to the agent.
Access settings
The access settings at the end of the tab allow you to define precisely who should be able to see and access the assistant.
Hidden: This toggle gives you the possibility to hide your assistant while you are setting it up or testing it. While this is toggled on, the assistant is only visible to the creator and managers (if set).
Managers: Through adding users as managers for an assistant you allow them to edit the settings. This is great if you want to collaborate with others on creating an assistant, without assigning them a permanent role (e.g. Global Moderator or Space Manager for team spaces). Deleting an agent is only possible for the creator, not managers.
Space: You can always adjust in which space an agent belongs. This allows you to quickly adjust the visibility and accessibility to a new user group.
Knowledge
In the "Knowledge" tab you can add relevant context to the agent. Think about which information and context is required by the agent to perform its task. For example this could be internal guidelines that should be considered, or existing documentation the agent should compare its research to.
The token usage always indicates how much of the context window is consumed by the knowledge you attached.
As long as you are below the limit all of the attached context is always loaded into the content for the AI model.
As soon as the limit is exceeded the agent automatically gets equipped with the "search company knowledge" capability, and will restrict this search to only the attached knowledge.
There are different sources from which you can add knowledge to your assistant:
From company knowledge
To add content that is already available on the platform, e.g. through the connected integrations. Two different options to access the company knowledge and find what you want to add:
Search: If you have a couple relevant content items, the assistant should consider and you know exactly which ones, you can enter a search term to find and select them.
Browse: Ideal if you know there is relevant context but don't know the exact name or it's not only single content items. It allows you to browse through your integrated systems and select the content on every hierarchy level (e.g. entire folders).
Tip: If there is a lot of relevant knowledge and not all of it in one folder, it's often useful to first create a Zive Collection in which you collect all relevant information about a topic, and then attach the collection to the agent. This brings a couple of benefits:
Easier to organize content in collections
Content can also be explored by users in the collection
Collection can be used in multiple assistants/agents potentially
Upload local files
If relevant files that are not indexed into the platform, but available on your device, you can quickly upload them directly into the agent's context.
Keep in mind that compared to content that you attach from the company knowledge, local files uploaded are static and not updated automatically. You should see this as a secondary option compared to the company knowledge.
Note: Local files uploaded into an agent become accessible for all users with access to this agent. Deleting an agent removes local files uploaded from the platform again.
Capabilities
Think of the agent as a work companion that performs tasks for you. The only thing you need to do is to equip it with the required tools.
Core capabilities
There are four default capabilities that you can activate for your agent:
Search company knowledge: This lets you define whether the agent should perform an internal search or not. Activate this if you expect the assistant to do research across the entire internal knowledge. On the other hand, if you expect the assistant to use very specific sources as context add them in the Knowledge tab instead and deactivate this capability to avoid the risk of the agent using other information.
Auto Pilot: Here you can define whether the agent should first return the sources qualified as relevant during the internal research and have the user confirm them (auto pilot off) or directly generate the answer without user confirmation of the relevant sources (auto pilot on).
Search web: Allows the agent to search relevant information from the web. Should be activated if the agent needs external, up-to-date information in order to perform its task.
Data analyst: This enables the agent to execute code, e.g. in order to create graphs or analyze data. Whenever you expect your agent to perform complex tasks or solve complex problems this capability will be required to achieve ideal results.
Image generation: If activated on platform level, assigning this capability enables the agent to create images as (part of the) output.
Tip: More is not always better! Less tools mean less options for the AI model to choose the correct one from.
Integrations
Note: Which integrations and respective actions are available depends on the configuration by your Admins.
In addition to the general capabilities you can activate specific integrations for your agents. This makes sense if you want the agent to interact with specific systems to retrieve additional information or create/update information.
You can simply click on "Add integration" to see a list of the available systems and actions to choose from. Select the actions the agent needs in order to perform its task. Below every action you can toggle on "Auto-run" to enable the action being performed without user confirmation.
Note: Without Auto-run enabled your agent cannot run fully autonomously because it requires user confirmation in the chat. We therefore recommend always turning it on!
Behavior
Last in order, but definitely not least. In the previous tabs you defined the framework conditions for your agent: the appearance, the context and the tools. In the "Behavior" tab you instruct your agent what you expect it to do and when.
Instruction
This field allows you to define a generic instruction for your agent. Usually this includes things like:
identity/role
task
expectations
rules
Model
Additionally you can select which model should be used by the agent.
Auto modus: The agent will decide automatically which of the models supported in the auto modus will be used
One specific model: You define the exact model for the agent.
Tip: Usually it makes sense to test which model delivers the best results and set this model as fixed. This leads to the most consistent results for all users of the agent.
Flow
Defining the flow is crucial for your agent. Here you configure when and in which interval the agent performs its process.
First you find the schedule settings, where you can choose between pre-configured options (hourly, daily, weekly) or set a custom interval if needed. This defines when your agent will run.
Afterwards you need to define what exactly should be done by the agent. Compared to Zive Assistants where this is optional it's required for Zive Agents, because they need a clear process to perform tasks without any additional user input.
How to update or delete an agent
Updating an agent
Of course, all of the above mentioned settings can be adjusted at any point in time. This is possible for all users with access to the settings which primarily depends on the space the agent is stored in:
Company space: Moderators (or managers that were set for the agent specifically)
Team space: Space managers (or managers that were set for the agent specifically)
Personal space: You (user that created the agent in his/her space)
Note: Keep in mind that agents in team spaces or the company space are likely activated by other users as well. Changes to agents do not only affect you, but also other users that have the agent activated.
Cloning an agent
Considering the note above, if you want to customize an agent that is available to other team members (team space) or your entire organization (company space), cloning can be helpful.
You find this option when you click on the three dots in the agent view.
By default the clone is stored in your personal space, where you can then configure it to your likings without the risk of affecting the behavior of the agent for other users.
Deleting an agent
You can simply delete an agent that you own after entering the agent, clicking on the menu icon and selecting delete. You also have this option via the same menu inside the agent settings.
Note: Same as for editing agents, keep in mind that agents in a team or the company space are likely activated by other users. Before deleting an agent it's worth checking if other users might still want to keep the agent.
How to use agents
Overview of your agents
Compared to Zive Assistants the overview always separates agents that you have activated (at the top) from agents that are available to you but currently not used by you (at the bottom).
If you have certain agents that you want to access frequently (e.g. to check the runs or adjust the settings) you have the option to pin them to your navbar.
Activate agents
If you want to start using an agent you can simply click on it. This will open a pop-up showing you the description and the schedule for the agent, to help you understand immediately what this agent will be doing and when.
Click on "Activate" and the agent will run for you according to its setup.
Check and deactivate your agents
When you click on one of your active agents you will always see an overview of all runs the agent performed since activation. This way you can check if all runs were completed smoothly or access runs from the past to take a look at the details again.
Additionally you have the option to
deactivate an agent, which will prevent any runs in the future or
trigger a run manually.
