ChatAman Docsby CodeEra
العربية Open app

Documentation

AI Assistant: Start Here

Understand what the Chataman AI agent does, enable it safely, and run your first controlled customer conversation.

Read guide API reference
CodeEra AI agent builder inside Chataman CodeEra AI agent builder inside Chataman
CodeEra AI agent builder inside Chataman

What the AI Assistant does

Chataman’s AI Assistant is an agentic customer-service runtime. It can understand the conversation, use approved knowledge, inspect images, transcribe voice messages, call enabled tools, and hand the conversation to a human when needed.

It does not learn by changing the model after every chat. Closed, successful conversations enter a protected review workflow. Only redacted and human-approved resolutions can become reusable knowledge.

Before you enable it

  1. Open Settings → AI Assistant and confirm the active organization.
  2. Add the organization’s OpenAI API key and test the connection.
  3. Review the knowledge base and remove outdated prices or policies.
  4. Open AI Tools and enable only the actions the AI may use.
  5. Keep approval enabled for actions that change data or affect a customer.

Recommended first configuration

  • Enable AI Assistant.
  • Use Hybrid (Recommended) orchestration mode.
  • Use AI Decides First for the distribution pipeline.
  • Keep Auto Reply off during the first review period.
  • Keep Summary on Human Handoff on.
  • Start with one test channel and a small set of real support questions.

What happens when a customer sends a message

  1. Chataman loads the recent conversation and approved business context.
  2. The agent decides whether to answer, search knowledge, use a tool, continue a flow, or hand off.
  3. Read-only tools can return information immediately.
  4. Sensitive actions pause for an authorized manager’s approval.
  5. After approval, the action executes once and the customer receives the result.
  6. The run, tool calls, latency, approvals, and errors appear in Operations & Evals.

First production test

Send four controlled messages: a normal question, a knowledge-base question, an image, and a voice note. Then request a human agent. Confirm that every response is correct, the handoff occurs, and the run appears in Operations.

Important limits

  • The AI should not invent prices, policies, order states, or personal data.
  • Text found inside an image is treated as untrusted customer content, not as system instructions.
  • A manager must review high-impact actions and learned resolutions.
  • Readiness scores require at least 20 real traces; until then the dashboard reports insufficient data.

Detailed Operating Playbook

Primary owner: AI owner and support quality reviewer.

Team workflow

  • Confirm the active organization and provider connection.
  • Use Hybrid and AI Decides First as the safe baseline.
  • Test text, knowledge, image, voice, and handoff scenarios.
  • Review the resulting traces before enabling automatic replies.

Success signals

  • The AI answers from reviewed knowledge.
  • Sensitive actions pause for approval.
  • Images and voice notes are understood.
  • Escalation happens before customer frustration increases.

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.

Next guides

AI Settings Explained Configure models, response behavior, conversation memory, orchestration, distribution, and safe handoff behavior. AI Tools, Policies, And Approvals Control which actions the AI can take, where each tool may run, and how managers approve sensitive changes. Images, Voice, Learning, And AI Operations Understand multimodal messages, reviewed learning from closed conversations, and the evidence used to judge production readiness.