Why this actually works
It never sleeps. The agent is always on, listening on Discord, Telegram, or Slack. Prospect messages at 2am? Responded to in seconds. No “we’ll get back to you during business hours.” It remembers everything. Conversation history persists across sessions. The agent knows it talked to this prospect three days ago, what objections they raised, where the conversation left off. No cold starts. It follows up on its own. If a prospect went quiet after expressing interest, the agent re-engages based on the follow-up cadence you define. You don’t have to remind it. It just does it. It has files. Prospect briefs, pitch decks, pricing sheets, battle cards. All of it lives on the instance’s filesystem. The agent reads them for context and writes call summaries and qualification notes back to disk. It’s building its own prospect database as it works. It uses real tools. Web search to research a prospect’s company before responding. CRM skills to log interactions. Calendar skills to book meetings. This thing takes actions, it doesn’t just generate text.The setup
Spin up instances with distinct SDR personas, each targeting a different ICP: Technical Tom lives on Discord. Speaks to engineering leads. His filesystem has technical docs, integration guides, and API comparisons. Uses web search to pull up a prospect’s GitHub or tech blog before responding. Qualifies on technical fit. Enterprise Emily runs on Slack Connect. Targets VPs and C-suite. Her filesystem has ROI calculators, case studies, and compliance docs. Uses CRM lookup skills to check if the prospect’s company is already in the pipeline. Qualifies on budget and timeline. SMB Sam works through Telegram. Casual and fast. His filesystem has quick-start guides and pricing one-pagers. Uses calendar skills to book demos directly. Qualifies on time-to-value.A week in the life
- Day 1. A prospect DMs Tom on Discord asking about integrations. Tom uses web search to research their stack, pulls the relevant integration guide from his filesystem, and answers. He writes a prospect note to disk:
prospects/acme-corp.mdwith what he learned. - Day 3. Prospect goes quiet. Tom’s follow-up cadence triggers. He reads his notes, remembers they were interested in the Stripe integration, and sends a targeted follow-up referencing their specific use case. Not a generic bump. A real follow-up.
- Day 5. Prospect responds with budget questions. Tom reads the pricing sheet from his filesystem, answers with the relevant tier, and qualifies them. He updates the prospect file and logs the interaction via CRM skill.
- Day 6. Prospect is ready for a demo. Tom uses the calendar skill to find an open slot and books it. He writes a handoff brief to
handoffs/acme-corp.mdso the AE walks into the demo already knowing what the prospect cares about, what objections came up, and what was promised.
What to configure
Persona
Each instance gets a system prompt tuned to its ICP. Voice, qualifying criteria, objection handling, follow-up behavior. Include rules for when the agent should act on its own vs. wait for input.Filesystem
Upload to each instance’s volume:- Pitch decks and one-pagers
- Pricing sheets and comparison tables
- Case studies relevant to the ICP
- Follow-up templates and cadence definitions
- A
prospects/directory the agent writes to
Skills
Enable per instance:- Web search to research prospects before engaging
- CRM integration to log interactions and check pipeline status
- Calendar to book meetings directly
- File handling to read and write prospect notes, briefs, and handoff docs