All agents

CRM System Agent

An always-on AI sales manager that scores every deal, sends or drafts personalized follow-ups on cadence, advances your pipeline, and logs everything — automatically.

View in marketplace Open dashboard

The CRM System Agent runs a single business's built-in CRM end to end. On each run it reads your live pipeline, enriches what it doesn't know, scores fit and intent against your ICP, decides the single next-best action per deal, acts through CRM tools (send or draft outreach, move stages, create tasks), and logs every step to the deal timeline. It works in three trigger modes — a scheduled pipeline sweep, an inbound-lead webhook, and inbound reply capture — and can run fully autonomously or human-in-the-loop.

What it does

This agent replaces the manual discipline of running a sales pipeline. It captures and dedupes leads into contacts and accounts, enriches accounts by fetching their website, and scores every deal 0–100 using a fixed weighting (0.5 ICP-fit + 0.3 intent + 0.2 engagement) with a one-line reason attached. For each deal due for a touch it calls a next-best-action brief, then takes exactly one action: send a personalized email (autonomous mode) or draft one and create a review task (human-in-the-loop mode), move the stage only when there's evidence, and reschedule the next follow-up. Outreach is constrained — match the configured tone, reference one specific detail, under 120 words, one CTA, signed as your configured sender — and replies to its emails are threaded back onto the deal automatically.

It operates entirely through tools (it never invents CRM state) and meters usage as it works (each AI action and new contact is recorded against the deployment). A daily sweep first processes any due multi-step sequences, then handles the highest-scoring stale deals (at most 5 per run); an inbound webhook captures and scores new leads; and reply capture detects intent and decides the next move. The built-in CRM is the system of record — no external CRM is required.

How it works

Your Business
1
Ingest
Reads the live situation: crm_list_pipeline returns stages, a weighted forecast, and the deals that are stale / due for follow-up now. On an inbound trigger it instead loads the new lead or the captured email reply.
2
Enrich
Fills gaps it doesn't know — crm_enrich_contact fetches the account's website and stores up to ~2,500 chars of readable page text on the account for future runs; web_search / web_fetch add public context.
3
Score
crm_score_deal sets a 0–100 score the agent computes as 0.5·ICP-fit + 0.3·intent + 0.2·engagement, with a one-line reason saved to the deal.
4
Decide
crm_next_best_action returns a compact brief (contact, account, score, recent activity, days stale) and the agent picks the SINGLE next-best action per deal: follow up, wait, advance, create a task, or close.
5
Act
In autonomous mode it sends a personalized email via crm_send_email; in human-in-the-loop mode it drafts the email and files a review task via crm_create_task. It moves the stage (crm_move_deal_stage) only where evidence justifies it. At most one outreach per deal per run.
6
Log & reschedule
Every email, note, and stage change lands on the deal timeline (crm_log_activity); crm_set_followup schedules the next touch (~cadence days out, defaulting to 3 if unspecified). Multi-step cadences are advanced by crm_process_sequences at the start of each sweep.
7
Report
Finishes with a single record_output call summarizing contacts touched, emails sent or drafted, deals moved, and tasks created.
Outcomes delivered

Setting it up — owner / admin

  1. 1
    Subscribe and deploy
    From the marketplace, subscribe to the CRM System Agent, then deploy it from its dashboard page at /dashboard/agents/[id]. No external CRM is required — the CRM is built in.
  2. 2
    Connect tools (mostly optional)
    Sending works out of the box via the platform mailer or your connected mailbox, with reply capture. The built-in CRM, web_search, and web_fetch need no setup.
  3. 3
    Fill the config fields
    Set Ideal customer profile (ICP) — drives scoring and messaging; Outreach tone (Friendly / Formal / Concise); Sender name and Reply-to / sender email; Follow-up cadence (Every 2 / 3 / 5 days or Weekly); Autonomy (Human-in-the-loop draft+task, or Autonomous send); Auto-reply to inbound leads (Off / On); and optional Pipeline stages (comma-separated, e.g. New, Qualified, Demo, Proposal).
  4. 4
    Set the schedule
    Choose Run frequency (Daily / Weekly / Monthly) — a daily sweep is typical. The inbound-lead webhook and email reply capture run in real time regardless of this schedule.
  5. 5
    Seed and first run
    Add a contact in the CRM console (/dashboard/crm) or POST a test lead to the inbound webhook, then hit Run now on the deployment page to watch the agent score the deal and send or draft a follow-up. Review the run summary.
  6. 6
    Tune autonomy
    Start in Human-in-the-loop to review drafts and tasks; once the messaging and scoring look right, switch Autonomy to Autonomous (and Auto-reply On) so it sends first-touch replies and follow-ups without you.

Using it day to day — your team

  1. 1
    Check the pipeline
    Open the CRM console at /dashboard/crm to see deals organized by stage, each with its AI score (0–100) and one-line reason, plus the weighted forecast.
  2. 2
    Work the review queue
    In Human-in-the-loop mode, the agent files tasks with drafted emails attached. Review each draft, edit if needed, and send — or let Autonomous mode handle sending for you.
  3. 3
    Follow the deal timeline
    Every outbound email, inbound reply, note, and stage change is logged automatically to each deal's timeline, so the full history is always current without manual data entry.
  4. 4
    Let replies drive the next step
    When a prospect replies, the reply is threaded back onto the deal as an email_in activity; the agent detects intent, updates score/stage if warranted, and either replies (autonomous) or creates a task for you.
  5. 5
    Read the per-run summary
    After each sweep, the run output reports contacts touched, emails sent or drafted, deals advanced, and tasks created — your snapshot of what moved.

Use cases

Autonomous follow-up
The daily sweep finds deals due for a touch (up to 5 per run, highest score first), writes a tailored follow-up, sends it (autonomous) or drafts a task (human-in-the-loop), and reschedules the next touch — so no lead is forgotten.
Inbound lead capture
Point your website/form webhook at the agent: it upserts the contact, creates a deal, enriches the company, scores it, and — only when auto-reply and autonomy are both on — sends a first-touch reply; otherwise it creates a task.
Pipeline hygiene
Deals are scored and moved only on evidence; stale opportunities resurface in the sweep instead of dying silently, and stages advance with a logged change.
Reply triage
Replies to the agent's emails are captured, intent-classified, logged as email_in, and turned into the next action automatically — a reply, a stage change, or a task.

What to expect

  • A self-maintaining pipeline with a 0–100 AI fit/intent/engagement score and one-line reason on every deal
  • Personalized follow-ups sent (autonomous) or drafted with a review task (human-in-the-loop) on your chosen cadence — at most one per deal per run
  • Inbound leads captured, enriched, scored, and optionally replied to via webhook
  • Every email, note, reply, and stage change logged automatically to a clean per-deal timeline
  • A per-run summary of contacts touched, emails sent/drafted, deals advanced, and tasks created, with usage metered per AI action and new contact

Metrics to watch

  • Follow-up coverage — share of due/stale deals actioned each sweep (the sweep handles up to 5 highest-score deals per run)
  • Email reply rate on the agent's outreach (replies are captured and threaded automatically)
  • Pipeline velocity — deals advancing stages on evidence vs. sitting stale
  • Lead response time — from inbound webhook to first-touch reply or task
  • Draft-to-send conversion in human-in-the-loop mode (how often you approve the agent's drafts) as a signal to move to autonomous
  • Metered usage — AI actions and contacts created per run, to keep spend predictable

FAQ

Will it email people without my approval?
Only in Autonomous mode. In Human-in-the-loop mode it drafts the email and creates a review task via crm_create_task instead of sending. For inbound leads it only auto-replies when Auto-reply is On AND Autonomy is Autonomous.
How does lead scoring work?
Each deal gets a 0–100 score the agent computes as 0.5·ICP-fit + 0.3·intent + 0.2·engagement, written via crm_score_deal with a one-line reason you can see on the deal. It rescores when new signal arrives (e.g. a reply).
Do I need an external CRM like HubSpot or Salesforce?
No — the built-in CRM is the system of record and requires no external setup. (HubSpot mirroring is offered by the separate Customer Acquisition agent, not this one.)
Does it capture replies?
Yes. Emails sent via crm_send_email are sent with a thread token so replies are captured back into the CRM automatically, threaded onto the deal as an email_in activity; that can trigger the next action.
How often does it run, and how many deals per run?
You set Run frequency to Daily, Weekly, or Monthly (daily is typical). Each sweep first processes due multi-step sequences, then handles up to 5 of the highest-scoring stale deals. Inbound webhooks and reply capture run in real time regardless of the schedule.