API configuration
- OpenAI API Key: secret used by this organization. Leaving the field empty keeps the saved key.
- Organization Key: optional OpenAI organization identifier when your account requires one.
- Model: model used for customer replies and agent decisions. Use the configured default unless you have tested another supported model.
- Max Tokens: maximum generated response size. Higher values may increase latency and cost.
- Temperature: creativity level. Customer support normally benefits from a lower, more consistent value.
- Timeout: how long Chataman waits for the provider before treating the request as failed.
Voice settings
- Voice transcription model: converts voice notes to text. The default is optimized for fast, cost-effective transcription.
- Speaker diarization: identifies different speakers in longer recordings. Enable it for calls or multi-speaker audio, not every short voice note.
AI features
- Chat Completion: allows generated conversational replies.
- Conversation History: includes recent messages so the reply follows the current context.
- Information Extraction: pulls structured facts from free text.
- Sentiment Analysis: detects tone and intent for routing and reporting.
- Content Moderation: helps block unsafe or policy-violating content.
- Auto Reply: allows automatic customer replies. Enable only after controlled testing.
- Auto Sentiment Analysis: analyzes every eligible incoming message.
Conversation and cost controls
- Max History Messages: number of recent messages passed to the model. More context can help, but increases token use.
- Cache Duration: how long reusable AI results remain cached.
- Confidence Threshold: minimum confidence expected before automated behavior is trusted.
- Daily Request Limit / Monthly Token Limit: spending and usage guardrails for the organization.
Orchestration modes
- AI First: AI makes the main decision and may invoke flows as sub-routines.
- Hybrid (Recommended): active matching flows continue; otherwise AI handles the conversation.
- Flow First: flows take priority and AI responds only when no flow is active.
Handoff and distribution
- AI Can Close Ticket: lets AI close a ticket when the goal is clearly complete. Keep off until your close rules are tested.
- Summary on Human Handoff: adds a short context summary for the receiving agent.
- Continue Active Flow First: lets an active flow finish its current path before AI takes over.
- Wait for AI / Flow Handoff: does not assign a human at conversation start.
- Distribute from the Start: legacy behavior that assigns a human immediately.
- AI Decides First: gives AI a clean opportunity to answer, run a flow, or escalate before human assignment.
- Max open chats per agent and Office Hours: protect the team from overload and out-of-hours assignment.
Safe baseline
Use Hybrid, AI Decides First, Summary on Human Handoff, a modest history limit, and explicit tool approvals. Change one setting at a time and verify it with a test conversation.
Detailed Operating Playbook
Primary owner: Organization admin or AI owner.
Team workflow
- Save and test the provider key.
- Choose the response and transcription models.
- Set conversation, usage, and timeout limits.
- Select orchestration and handoff behavior, then test one setting change at a time.
Success signals
- The connection test succeeds.
- The selected model is shown in settings.
- Usage limits are documented.
- A test conversation follows the selected orchestration mode.
When to review or escalate
Review this workflow when channels, policies, or workload change. Escalate to a manager when numbers look inconsistent or the same issue repeats more than once in the same day.