sales-agent-lab

Voice-first, human-in-the-loop AI sales agent lab for Tectara Systems / Webton.

Connects to Kundenportal (leads + drafts + approvals) and MailOps (email send + replies) without changing any Webton AIO source code.

Quick start

1. Copy .env.example to .env and fill in the service tokens Kenny provides.

2. Create a virtual environment and install dependencies:

   uv sync
   # or
   python -m venv .venv && source .venv/bin/activate && pip install -e .

3. Run the agent CLI:

   sales-agent --help

M0: Scaffold (current)

• Project structure
• Kundenportal + MailOps clients
• Agent configuration
• ADOS sales knowledge (discovery script + lead scorecard)
• .env.example with all required variables

M1: Text-first outbound drafter

• Poll Kundenportal for leads with status=new and min_quality.
• Qualify leads using the ADOS scorecard via LLM structured output.
• Generate personalized German outreach emails.
• Post drafts to Kundenportal with requires_approval: true.

M2: Approval → send

• Receive draft.approved events from Kundenportal.
• Trigger MailOps to send approved emails.
• Update lead status in Kundenportal.

M3: Inbound replies

• React to inbound email replies (MailOps webhook or IMAP poll).
• Draft reply in Kundenportal for approval.

M4: Voice channel

• Add Pipecat or LiveKit voice transport around the same agent core.
• Streaming ASR + VAD + streaming TTS.

Repository

https://github.com/imKXNNY/sales-agent-lab