This page is different from the Chat and Email tabs.
The settings below affect the agent’s core identity and response behavior.
The settings below affect the agent’s core identity and response behavior.
General settings
Name
Your internal display name for this agent.- Shows in the Agents sidebar and settings screens
- Helps teammates identify the right agent quickly
- Use role-based names (
Support Assistant,Billing Assistant) - Avoid ambiguous names (
Agent 1,Test Bot)
Avatar
Profile image used for this agent in UI previews and agent lists.- Purely visual/identity
- Does not change model behavior
- Use a clear, brand-aligned icon
- Keep style consistent across all agents in a workspace
ID
Read-only unique ID for the agent, with copy-to-clipboard.- Useful for support/debugging
- Safe to share internally when referencing a specific agent config
Guidance settings (behavior controls)
These are the most important controls for output quality.Tone
Controls how the agent says things. Available options:WarmNeutralProfessionalDirectWitty
When to use each tone
| Tone | Best for | Watch out for |
|---|---|---|
| Warm | Customer support, high-empathy moments | Can become verbose or too casual |
| Neutral | Mixed workloads, general support | Can feel generic if overused |
| Professional | B2B, compliance-heavy, formal comms | Can sound rigid if over-formal |
| Direct | High-volume queues, urgent troubleshooting | Can sound abrupt if not softened |
| Witty | Brand voice with personality | Can sound flippant in sensitive cases |
Tone best practices
- Choose one tone per agent and keep it stable
- Match tone to audience (B2B vs consumer)
- If you use Witty, add clear prohibited phrases and context guardrails
Response length
Controls how much detail the agent gives. Available options:- Short
- Standard
- Long
How to choose
| Length | Best for | Risk |
|---|---|---|
| Short | High-volume support, quick confirmations | May omit required detail |
| Standard | Most support teams (default recommendation) | Low risk |
| Long | Complex technical issues, policy-heavy answers | Can overwhelm customers |
Prohibited phrases
A managed list of words/phrases the agent should avoid using when generating responses. Examples:- “That’s our policy”
- “I can’t help with that”
- “As an AI language model”
How it works
- You maintain the list by clicking Manage
- The system instructs the model to never use these phrases
Context
Business context that helps the agent understand your business.What to include
- Product/service scope
- Support boundaries (what you do and don’t support)
- Contact and escalation paths
- Key policies (refunds, SLAs, shipping windows)
- Canonical links or docs to trust
What not to include
- Secrets (API keys, passwords, private credentials)
- Time-sensitive data likely to go stale quickly
- Ambiguous or conflicting policy text
Good context example
Weav is a B2B customer support platform for SMB teams. Support hours: Mon–Fri, 9am–6pm ET. Refund policy: monthly plans can be refunded within 14 days of first payment. For billing disputes, route customers to billing@weav.com and ask for invoice ID. Do not provide legal advice or respond to questions unrelated to Weav. For security requests, escalate to security@weav.com.
Bad context example
We are the best support company. Be helpful and nice. We do many things.
Context best practices
- Write for decisions, not marketing
- Use short bullets and explicit rules
- Update when policy/process changes
- Add examples of edge cases your team sees often
How these settings work together
At generation time, the agent is guided by:- Base system prompt
- Your Context
- Your Response length
- Your Tone
- Your Prohibited phrases