What Connected Apps Can Do
A connected AI app works with the same agents, sessions and metrics you see in the Admin app. What it can do depends on the access you chose when approving the connection: Read Only apps can look but not change anything, and Read and Write apps can also build and manage agents. Some actions are never available to a connected app, whatever its access. The connect guide summarises the same lists under What It Can Do.
Before You Start
- Connect an app first. See Connecting an AI App.
- The app acts as you. It can't reach anything your own role can't, and it covers every environment in the organisation you connected.
What Every Connected App Can Read
Every connection, including Read Only ones, can:
- list environments and whether each one is ready to use
- list agents, with their stats for the last 30 days
- read an agent's configuration and tools, with secrets removed
- list an agent's published versions, read any one of them, and compare two versions or a version with the draft
- read metrics for any date range
- read sessions, with their scores and summaries
- read full transcripts and tool calls
- list phone numbers
- read agent blueprints and guidance on building agents
What Apps With Build Access Can Do
Read and Write connections can also:
- preview and check a draft agent before it is created
- create agents
- update agents
- enable or disable agents
- test an agent by sending it a message
When an app updates an existing agent, the change is saved to the agent's draft, the same draft the agent editor works on. A person still publishes it from the Admin app before callers get the new version. See draft and live.
What Connected Apps Can't Do
No connected app can:
- delete agents
- change Twilio, routing, wrap or capacity settings
- see secrets or API keys
These stay in the Admin app, with the people whose role allows them. See Settings for the settings an Organisation Admin manages.
Prompts to Try
The Prompts to Try page of the connect guide has prompts you can copy and paste into your AI app once it is connected. The guide fills in the name of your current environment and agent where a prompt needs one. Below, they appear with the placeholders "Production" and "your agent". Some prompts make the AI app show a view, such as an agents overview or a metrics dashboard, as well as a written answer.
Explore
- "Show me all agents in Production and how they did this month". Shows an agents overview.
- "Which agent had the most sessions last week?"
Diagnose
- "Why did your agent's score drop? Look at the lowest-scoring sessions". Shows a metrics dashboard.
- "Show me the transcript of your agent's worst session this week". Shows the transcript.
Build
These need Read and Write access.
- "Help me design a voice agent that books appointments and hands off to a human when unsure". Shows a draft agent you can create.
- "Create a chat agent from this FAQ", with your FAQ pasted after it. Shows a draft agent you can create.
Improve
- "Review your agent's prompt against its recent transcripts and suggest changes"
- "Test your agent with a caller who wants a refund". Testing needs Read and Write access.
Replace "your agent" with the name of one of your agents, for example "Why did ACME Reception's score drop?".
What Happens Next
Answers come from your data at the moment you ask. Agents an app creates or changes show up in the Admin app straight away. Updates to an existing agent wait in its draft until someone publishes them.
Related
- Approving Access. Choosing Read Only or Read and Write
- Managing Connections. Disconnecting or revoking an app
- Sessions. The sessions and transcripts an app reads