Agents
Describe the job and Evose builds the agent
Say who it serves, what it does and when to hand off to a person. Evose writes the prompt and connects knowledge, tools and skills. Test it, then publish it to your team.
Make a support agent that answers customer questions about plans and refunds. Check the subscription first, and send refunds over $500 to Omar.
Give it one clear job
Build it by talking
No code and no blank form to start from. Describe the job and the Builder writes the prompt and picks the knowledge and tools it needs.
Try it, then publish
Test the draft in preview as often as you like. After you publish, editing the draft leaves the live version alone.
Your team and Evose AI can call it
People can chat with it directly, projects and schedules can assign it work, and Evose AI can hand it part of a task.
From a description to a live agent
Three steps to an agent your team can use.
- 01
Describe the job
Say who it serves, what it does and when to hand off to a person. The Builder picks a name and avatar, writes the prompt and connects knowledge and tools.
- 02
Try it in preview
Ask it real questions and see which sources it read and which tools it called. If something's off, say so and the Builder updates the draft.
- 03
Publish to your team
Choose who can use it and publish a version. Later edits stay in the draft until you publish again.
Ways to use it
Customer support
Answers customers from the help center after checking the order.
Prompt “Make a support agent that answers plan and refund questions from the help center and checks the order with Subscription lookup first. Don't promise refunds over $500, pass those to Omar.”
Sales practice
Plays the buyer for new reps and scores them against the playbook.
Prompt “Make a sales practice partner that plays an enterprise buyer comparing options and scores each session against the playbook. Name only the one thing to fix and don't write the rep's lines for them.”
Weekly report
Reads this week's finished tasks and writes the team report.
Prompt “Make a report agent that reads the tasks finished in our project this week and writes one page using the Weekly report skill. Flag numbers that don't match, and show me before it goes to the team.”
What an agent is made of
Ask the Builder to change any part in chat, or open the configuration and change it yourself.
Two ways to build
A prompt agent follows one set of instructions and suits a focused assistant. An Orchestrated Agent splits the conversation into nodes and branches for flows that go step by step.
Knowledge
Connect knowledge bases and choose smart use, retrieval only or document reading. Access follows each knowledge base's own permissions.
Tools and skills
MCP servers, plugins and marketplace tools let it look things up and send messages. Skills are your team's written methods, loaded when they apply.
Calls other agents and workflows
It can hand part of a task to another agent or run a workflow, then carry on with the result.
Memory
When it's on, the agent keeps the main points of past conversations for next time. Members can view, edit or delete those memories, or turn memory off.
Images and files
It can read images and files shared in chat and use the image, audio and video models your organization has connected.
Always know which version is live
Drafts, versions and permissions are managed separately, so your team always uses a published version.
- Drafts don't touch the live versionWhile you edit, people keep using the published version. Changes go live when you publish.
- Go back to an earlier versionSwitch to any published version from the history, or copy an old version into the draft and keep editing.
- Resources follow permissionsThe Builder only connects knowledge and tools you're allowed to use. Workspace roles and resource permissions decide who can view, use or edit.