AI Agent

AI Customer Service Agent

An autonomous AI agent that runs as a member of your team — not a chatbot. It reads from your stack, acts under explicit permissions, and reports outcomes to humans.

What this agent does

  • Resolve common questions without human handoff
  • Escalate edge cases to the right team member with full context
  • Generate per-ticket summaries for QA review
  • Detect sentiment shifts and flag at-risk customers

How it works

  1. 1

    Subscribe & connect

    Install the AI Customer Service Agent agent in your WorkAist instance. Authorize the connectors it needs.

  2. 2

    Configure boundaries

    Set permission isolation (which tables, which actions). Define escalation rules for humans.

  3. 3

    Run

    Trigger by cron, webhook, button, or another agent's event. The Process Map shows every step live, including token and dollar costs.

  4. 4

    Review

    Audit log captures every action. Impact Timeline shows what changed and why. Iterate the agent's CLAUDE.md to refine behavior.

Pricing

WorkAist is open-source under AGPL-3.0 — the agent itself is free. You pay only for:

  • LLM tokens consumed by the agent (your Anthropic / OpenAI key)
  • The underlying connector services (if they charge for API access)

Self-hosted on your own infrastructure (Docker / Hetzner) — no per-seat lock-in.

Trust & compliance

Reversible PII redaction

Sensitive values are tokenized before prompts and restored on the response. The LLM never sees raw PII.

Permission isolation

Group → agent → table-level ACL. The agent only sees data its permissions allow.

Audit log

Every read, write, and decision is logged with timestamp, source, and approval chain.

AGPL-3.0

Open-source license, self-hostable. Same model as GitLab CE. Source code is fully inspectable.

FAQ

Is AI Customer Service Agent truly autonomous, or does a human have to approve every step?

Both modes are supported. By default AI Customer Service Agent runs autonomously within the boundaries you set; any action that crosses a defined risk threshold (large invoice, customer-facing email, irreversible operation) requires a human approval via Telegram or the web UI.

What if AI Customer Service Agent makes a mistake?

Every action is logged. Impact Timeline shows what changed and why. Most actions are reversible (rollback in 1 click). For irreversible actions, the agent always asks for human approval first.

Can I customise AI Customer Service Agent's behavior for my company?

Yes. Each agent is a folder with a CLAUDE.md file describing its role, tone, and rules in plain language. Edit it in the browser-based VS Code (with Claude Code CLI built in) and changes take effect on the next run.

Does AI Customer Service Agent replace humans?

No. It handles the repetitive and the predictable, freeing humans for judgment calls, relationships, and edge cases. Most customers see AI Customer Service Agent as a "force multiplier" rather than a replacement.

How do I integrate AI Customer Service Agent with my existing tools?

AI Customer Service Agent uses MCP (Model Context Protocol) connectors. WorkAist ships with 1,300+ connectors covering accounting, CRM, banking, e-commerce, communication, and more. Custom integrations take a few hours via the AI Wizard.

Deploy AI Customer Service Agent today

One agent, one CLAUDE.md file, your data stays yours. Open-source under AGPL-3.0.

Get started
AI Customer Service Agent — autonomous AI agent for your business | WorkAist