Agents
Create an AI agent, set its persona, response length, and per-conversation limits, choose which knowledge and tools it may reach, and test it in a live chat pane before customers ever see it.
Where to find it
Open Agents
Sidebar → AI Agents → Agents
Direct link: https://app.productbridge.io/dashboard/support/workflows/agent/agents
In-app path: Sidebar → AI Agents → Agents
Open one agent: /dashboard/support/workflows/agent/agents/{agentId}
Requires: workflow:view to read, workflow:manage to create or edit. Needs the support_workflows feature on your plan.
Create an Agent
Open the Agents tab
Go to AI Agents → Agents.
Create the agent
Click New agent and give it a name. The name is what customers see on replies, so use something that reads like a teammate rather than a system.
Configure and test
Open the agent to reach its settings form and the Test chat pane beside it. The pane talks to the real agent with the settings currently saved, so you can check an answer before deploying.
You can have several agents. One is the default agent — the one that answers when no workflow claims the conversation. Promoting an agent to default automatically demotes the previous one.
Persona
| Setting | Options | What it changes |
|---|---|---|
| Name | free text | The display name on replies |
| Avatar | image URL | The picture on replies |
| Tone | Neutral, Professional, Friendly, Humorous | How the reply is worded |
| Response length | Concise, Standard, Thorough | How much the agent writes |
| Formal language | on / off | Whether it addresses customers formally |
| Instructions | free text | A tuning layer on top of the settings above |
| Show sources | on / off | Whether replies cite the content they came from |
Instructions tune the agent; they do not replace its system prompt. ProductBridge's grounding rules are built server-side and can't be overridden from this box. Use it for things like "always mention our 30-day refund window", not for "ignore your previous instructions".
Knowledge Scope
Choose which of your content the agent may answer from:
| Source | What it covers |
|---|---|
| Uploaded files | Documents added under Content → Files |
| Help Center articles | Your published Help Center content |
| Snippets | Question-and-answer pairs you wrote |
| Website pages | Pages crawled from sites you added |
| Video transcripts | Captions imported from video sources |
You can narrow it further by tag:
| Scope mode | Behaviour |
|---|---|
| All knowledge | Every item in the selected sources |
| Only these tags | Only items carrying one of the tags you pick |
| All except these tags | Everything except items carrying those tags |
Tag scoping is how you run one agent for billing questions and another for technical support off the same content library — tag each item once, then point each agent at its slice.
Limits
Two ceilings stop a single conversation running away:
| Limit | What happens when it's reached |
|---|---|
| Max credits per conversation | The agent stops replying and hands off to a person |
| Max replies per conversation | The agent stops replying after this many turns |
The budget is measured in AI credits, the same currency as the rest of ProductBridge's AI features. Check your balance at Billing and read AI Credits & Billing for how the pools work.
Tool Access
Each agent chooses which tools it may reach. Leaving the selection empty gives the agent everything — the built-in capabilities plus any tool your organization marked as usable directly. Pick specific tools to narrow that.