> ## Documentation Index
> Fetch the complete documentation index at: https://docs.reply.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Build a supervised AI SDR

> Give Jason a goal, knowledge, and guardrails — then supervise through approval gates instead of operating by hand.

**Outcome:** a supervised, self-improving outbound system — Jason (Reply's AI SDR) writes and sends
outreach and handles replies, while your agent supervises through approval queues and performance
data. The highest-altitude pattern: an agent configuring an agent.

## Best interface

<Note>
  **MCP** — the Jason toolset is richest there: 31 of the 70 tools drive knowledge bases, offers,
  playbooks, reply handlers, reengagement cards, and the approval queue. The REST API covers the
  knowledge/offer/playbook layer.
</Note>

## Steps

<Steps>
  <Step title="Define the sales goal — available">
    Translate the objective into an offer, a target market, and success metrics. Draft offer
    content from a website with MCP `reply_generate_offer_from_website` (review, then
    `reply_create_offer`).
  </Step>

  <Step title="Build the knowledge base — available">
    `reply_create_knowledge_base` → add URL sources (`reply_add_knowledge_base_source`), reply
    handlers (how Jason answers pricing/security/… questions), and reengagement cards (win-back
    plays). See [MCP tools → Jason AI SDR](/mcp/tools#jason-ai-sdr).
  </Step>

  <Step title="Define offers and playbooks — available">
    The offer says *what's being sold and to whom*; the playbook says *how* (tone, structure,
    channels). `reply_create_offer`, `reply_create_playbook`, then attach both to the sequence.
  </Step>

  <Step title="Point it at prospects — partly coming soon">
    Enroll contacts you already have ([Launch outreach](/guides/launch-outreach)).
    <Warning>Intent-signal targeting and Live Data sourcing are **coming soon** — see [Find prospects](/guides/find-ideal-prospects).</Warning>
  </Step>

  <Step title="Set the reply mode — available, gate it">
    `reply_set_sequence_reply_mode` → **Review** (Jason drafts, you approve) is the safe default.
    **Autonomous** (Jason sends without approval) is a [high-stakes switch](/agents/safety) —
    verify offer/playbook/knowledge base first and confirm explicitly. Recipe:
    [Switch Jason to Autonomous](/mcp/recipes#switch-jason-to-autonomous).
  </Step>

  <Step title="Supervise the queue — available">
    `reply_list_pending_approvals` → approve (sends immediately), regenerate with feedback, or
    reject (removes the contact from the sequence — irreversible). Recipe:
    [Supervise Jason in Review mode](/mcp/recipes#supervise-jason-in-review-mode).
  </Step>

  <Step title="Watch conversations and optimize — available">
    Replies flow into the unified inbox with AI categorization; escalate high-intent threads to a
    human. Read the meetings ledger and per-sequence stats; refine the offer, playbook, and
    targeting based on what converts ([Analyze performance](/guides/analyze-performance)).
  </Step>
</Steps>

## Availability notes

* Knowledge bases, reply handlers, reengagement cards, offers, playbooks, reply-mode control, and
  the approval queue: **available today via MCP** (and the AI SDR REST module).
* **Coming soon:** intent-signal prospect targeting; full autopilot enrollment loops. Everything
  labeled available above is callable now.

## Related

* [MCP recipes: Supervise Jason / Switch to Autonomous](/mcp/recipes#supervise-jason-in-review-mode)
* [Convert replies into meetings](/guides/convert-replies-to-meetings) — the human-approved counterpart
* [Agent safety rules](/agents/safety) — why Autonomous is gated