> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cekura.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Cekura Agent
> Chat with an AI assistant that can run Cekura operations on your behalf — create agents, generate scenarios, design metrics, review results, and more.
## What is Cekura Agent?
Cekura Agent is an AI assistant embedded in your Cekura workspace. You describe what you want in plain English and the agent performs the action — using the same Cekura APIs and skills you'd use yourself, but without you having to click through the UI or remember endpoint names.
It's designed for power users who want to move faster on tasks like:
* Creating or updating AI voice agents
* Generating evaluation scenarios
* Designing and previewing metrics
* Triggering test runs
* Exploring your workspace data (agents, scenarios, metrics, calls, results)
Cekura Agent is currently in **Beta**. Destructive operations (deleting
agents, scenarios, metrics, etc.) are disabled for safety — perform deletes
in the regular Cekura UI.
## Getting started
Click **Cekura Agent** in the left sidebar under the *General* section.
The URL is `/ai-assistant`.
Type your request in plain English — e.g. *"Show me all scenarios for agent 42"* or *"Create 5 scenarios for my agent that test refund edge cases"*.
When the agent is about to create or change something, it shows a
confirmation with the exact parameters. Reply **yes** to proceed or **no**
to cancel.
The agent remembers the whole chat. Past sessions show up in the left pane
so you can revisit them.
When generating evaluators, if you expand **Additional Configuration** during
the creation flow, tools are **not** included by default — evaluators will be
created without the End Call tool, which causes runs to stall. After
generation, verify each evaluator has the `end_call` tool. To add it in bulk:
go to **Evaluators**, select the affected evaluators, and use **Actions →
Modify Evaluators → Add Tools → End Call**.
## Example prompts
* "List all agents in this project"
* "Show me the metrics for agent 12"
* "How many scenarios does agent 3 have?"
* "What did the last run for agent 7 look like?"
* "Generate 5 scenarios for agent 42 covering refund edge cases"
* "Create a metric that checks whether the agent asked for the customer's date of birth"
* "Add a scenario where the user is frustrated and wants to escalate"
* "Read my requirements.txt and generate one scenario per requirement"
* "Use this CSV to create scenarios matching each row"
* "Run the evaluators for agent 5"
* "Show me the results from the last run"
* "What metrics failed most often?"
## Features
### Quick-action suggestions
When you open a new chat, four suggestion cards appear. Click any of them to
seed the input with a useful starting prompt — a fast way to discover what the
agent can do.
### File attachments
Attach context files directly in the chat using the paperclip icon.
* **Max 5 files per message**
* **Max 2 MB per file**
* **Supported formats:** `.txt`, `.md`, `.csv`, `.json`, `.yaml`, `.yml`, `.xml`
The agent reads file contents and can use them directly — for example,
generating scenarios based on a spec document, or creating metrics from a
checklist. Binary formats (PDF, images, audio) aren't supported yet.
### Clarifying questions
If your request is ambiguous — for example, *"List the scenarios"* without
specifying an agent — the agent asks follow-up questions inline. Answer each
one and the conversation continues naturally.
### Confirmation before writes
Before creating or updating anything (an agent, a metric, a scenario), Cekura
Agent shows the exact parameters it's about to send and asks you to confirm:
> *Confirm: I am about to call `scenarios_create` with:*
>
> * *agent\_id: 42*
> * *name: "Refund edge case"*
> * *personality: 5*
>
> *Reply 'yes' to proceed or 'no' to cancel.*
This lets you review changes before they happen. For update operations, the
confirmation shows a diff of each changed field (`old value → new value`) so
you can see exactly what is changing.
### Chat history
Every conversation is saved and appears in the left sidebar. Click any past
chat to reopen it. Delete a session with the trash icon.
### Stop a generation
If the agent is taking too long or heading in the wrong direction, click the
**Stop** button to cancel the current response. Your message and partial
response are preserved; you can follow up with a clarification or a new prompt.
### Custom skills
Beyond its built-in skills, the Cekura Agent can be extended with your organization's own **custom skills**, so it can run org-specific workflows on your behalf. Custom skills are org-scoped. Once added, they're available to the Agent for everyone in your org, and any member can view the active list.
Add a skill as a **source**: either a **single file**, a **GitHub repository** (provide the repo URL), or an **uploaded archive** (`.zip` / `.tar.gz`). You can then **enable, disable, or delete** individual skills, and **re-sync** a GitHub source to pick up changes you've pushed.
To **add custom skills**, navigate to **Settings** → **Custom Skills** in the UI.
## Limitations
* **No delete operations.** The agent cannot delete agents, scenarios,
metrics, call logs, cron jobs, projects, or test profiles. Use the regular
Cekura UI for deletes.
* **No dashboard editing.** Creating, updating, or deleting dashboard widgets
is not currently supported via Cekura Agent. Use the regular Cekura UI to
manage your dashboards and widgets.
* **Text files only.** Attachments must be one of the supported text formats.
* **Session context is capped.** Very long conversations are truncated to
roughly the last \~40,000 characters of chat history per request — the
agent may lose track of early messages in extremely long sessions. Start a
new chat when context grows stale.
* **Beta feature.** Behaviour may change as we iterate. Please
[share feedback](mailto:support@cekura.ai) if you run into rough edges.
## Troubleshooting
This can happen when a tool returns an unusually large response and the
agent is processing it. Click **Stop**, refresh the page, and try
narrowing your request — e.g. filter by agent ID, folder, or tag instead
of asking for everything at once.
Sessions idle out after 10 minutes without activity. A new session starts
automatically on your next message, and the chat history is preserved in
the sidebar.
That's intentional — destructive operations are disabled for safety.
Perform deletes in the regular Cekura UI.
Your message history is preserved. Just type a follow-up or restart the
original request — the agent has the full conversation context.
When generating evaluators, if you expand **Additional Configuration** to
enter parameters manually, tools are not automatically included — the
evaluator is created with no tools enabled by default. This means the
testing agent cannot hang up the call, which can cause runs to stall.
**To add the end call tool to existing evaluators in bulk:**
1. Go to **Evaluators** in the Cekura dashboard.
2. Select all affected evaluators using the checkbox.
3. Click **Actions** → **Modify Evaluators** → **Add Tools**.
4. Add the **End Call** tool and save.
For new evaluators, verify the **Tool Calls** section includes the `end_call`
tool before triggering runs.