Active Agents
8 built · 8 live
📤
Outreach Agent
Cold Email Engine
Sends personalized cold emails to independent medical practices 3x daily. Targets office managers & practice owners. Pulls from prospect list, queues via Make.com.
/health
/run
💼
Ava
LinkedIn Content Agent · #6
Posts to your LinkedIn 4x per day. Rotates through AutomatedRCM promos, AI in healthcare, billing tips, and founder story. Never repeats — uses KV memory to track post history.
/debug
/history
📥
Rachel
Reply Detection Agent · #2 Agentic
Scans Gmail every 15 min. Claude classifies replies as INTERESTED / NOT_INTERESTED / QUESTION / OUT_OF_OFFICE. Auto-responds to warm leads, then alerts Laureen on hot leads via both email and Twilio SMS — in parallel.
/health
/run
/debug
/history
/test-sms
🌅
Morgan
🪐 Hub
Morning Manager · #7
7 AM daily briefing emailed to you. Chat with Morgan and she speaks for the whole team — every /chat call queries Outreach, Ava, Rachel, Faye, Sentinel, Greta, and Finn in parallel via service bindings, then synthesizes one unified answer. Ask "how's everything going?" and you get the throughline across operations, pipeline, security, grants, and funding — not 7 separate updates.
/health
/run
/debug
/history
📬
Faye
Follow-Up Agent · #9
Runs every morning at 8 AM ET. Reads the shared LEAD_KV, finds WARM/INTERESTED leads not contacted in 3+ days, drafts a personal follow-up with Claude Haiku, sends it via Gmail, and emails Laureen a per-lead summary. Caps at 25 sends/run and 4 follow-ups per lead.
/health
/run
/debug
/candidates
/history
🛡️
Sentinel
Security & Health Monitor
Daily at 9 AM ET: pings every live worker via service bindings, emails Laureen if any return non-2xx. Sundays: weekly credential audit — exercises Anthropic, Gmail OAuth, Twilio, and LinkedIn tokens and always emails an all-clear or failure summary.
/health
/run
/audit
/debug
/history
💰
Greta
Grants Agent · weekly scout
Runs every Monday morning at 7 AM ET. Scans a curated list of 20 verified no-fee grant programs (NIH SBIR, SBA, Amber Grant, Cartier Women's Initiative, Hello Alice, Florida WBC, etc.), filters by current LLC/EIN status + 90-day deadline window, and emails Laureen a ranked weekly grant report. Never invents programs. Never submits — surfacing only.
/health
/run
/opportunities
/debug
/history
🚀
Finn
Funding Agent · NO LOANS
Runs every Monday morning at 7 AM ET. Scans a curated list of 20 verified non-loan funding programs — equity-free accelerators (NVIDIA, Microsoft, Google, AWS), equity accelerators (YC, Techstars, MedTech Innovator), pitch competitions (Pioneer, Visa), and women-focused VC (Backstage, FFF). Honors Laureen's no-loans rule at both the curated-list level and in Claude's reasoning.
/health
/run
/opportunities
/debug
/history
🧠
LEAD_KV — Shared Memory
Prospect Intelligence Layer · Level 3
KV-based persistent memory per prospect, keyed by email. Outreach writes COLD on every send; Rachel escalates to WARM/INTERESTED on replies; Faye bumps emailsSent on follow-ups; Morgan reads the funnel for daily briefings. Status only escalates — never demotes.
Full Autonomous RCM Pipeline
9 stages · 3 live · 6 planned
This is the complete autonomous RCM cycle — from a new patient walking in the door to money in the practice's bank. Each stage can have a dedicated AI agent. You already own the top of the funnel (outreach). Here's the full picture:
Outreach
Outreach Agent
Reply Mgmt
Rachel
Eligibility
Vera (planned)
Prior Auth
Availity Bot
Coding
Coda (planned)
Claims
Clara (planned)
Denials
Dex (planned)
Payments
Penny (planned)
Reporting
Morgan
✅ What You Own Right Now
You control the top of the funnel: finding practices, sending outreach, managing replies. Ava builds your brand so practices already know you when they get Rachel's reply. This is Stages 1–2 of a 9-stage autonomous pipeline.
🔮 What Full Autonomy Looks Like
A practice signs a contract → agents handle everything: check insurance, submit claims, catch denials, post payments, alert you only on exceptions. You review dashboards. Agents do the work. This is the business you're building.
⚡ The No-Availity Start (Smart Move)
You're right that some stages need Availity (eligibility checks, prior auths) which requires provider credentials. You can build everything else first — outreach, reply handling, intake forms, coding assistance, denial analysis — and plug Availity in when you land your first client. The agents don't have to wait for it. Start where you have access, layer in integrations as clients bring their credentials.
The 5 Levels of AI Autonomy
Your system today: Level 1–2
Most people building "AI automation" are at Level 1. What you're building is a Level 3-4 system that will eventually reach Level 5. Here's exactly what each level means and where your agents sit today:
1
Scheduled Automation — Runs on a clock, follows a script
No decisions. No memory. Same action every time. Like a cron job with AI-written content. Claude writes it once, the worker runs it forever. YOU don't need to be there.
✅ Outreach Agent
✅ Ava (LinkedIn)
✅ Email Briefing
2
Reactive Agent — Reads inputs, classifies, responds differently
The agent looks at what came in, makes a decision (INTERESTED vs NOT_INTERESTED), and responds accordingly. It's not just executing — it's thinking about what to do next.
✅ Rachel (Reply Agent)
✅ Morgan (Morning Manager)
3
Memory Agent — Remembers every contact, every interaction
The agent knows what it sent before, what the person replied, what stage they're in. It doesn't repeat itself. It builds a relationship over time. This is where it starts feeling like having a real employee.
🔮 Memory System (planned)
KV-based contact memory
4
Instruction-Following Agent — You talk to it, it changes behavior
You text "Ava, stop posting about billing for this week, focus on denial management." The agent hears you, updates its behavior, and confirms. This is what you're asking about — talking directly to the agent. This requires a /chat endpoint + SMS or dashboard bridge.
🔜 SMS Bridge (Twilio)
🔜 Dashboard /chat endpoint
5
Learning Agent — Tracks what works, adjusts its own behavior
The agent monitors its own reply rates, open rates, conversion rates. It tests different subject lines and body copy. It learns "Tuesday mornings get 3x more replies" and shifts its schedule automatically. This is full autonomy.
🔮 A/B testing layer
🔮 Self-optimizing loops
💡 How "Talking Directly to Ava" Actually Works
Right now, Ava is a Cloudflare Worker with a cron schedule. She runs, posts, and goes back to sleep. She doesn't "listen."
To talk to her, you need to add a
The chat panel on the right is showing you exactly what that would look like. The responses are simulated now — but when the /chat endpoints are deployed, it becomes real two-way communication with your agents.
To talk to her, you need to add a
/chat endpoint to her worker. When you send a message there, the worker calls Claude with Ava's full system prompt + your message + her post history, and returns a response. Then you wire that to a form in this dashboard (like the sidebar on the right).The chat panel on the right is showing you exactly what that would look like. The responses are simulated now — but when the /chat endpoints are deployed, it becomes real two-way communication with your agents.
Build Roadmap
Next session priorities
Deploy Rachel — DONE 2026-05-25
Worker automatedrcm-reply-agent live at https://automatedrcm-reply-agent.laureennicholson635.workers.dev with REPLY_KV namespace, */15 cron, all 5 secrets set, and a noreply sender filter. End-to-end verified via /debug.
Live
Morgan — Morning Manager — DONE 2026-05-25
Worker automatedrcm-morgan live at https://automatedrcm-morgan.laureennicholson635.workers.dev with MORGAN_KV, cron 0 11 * * * (7AM ET), all 5 secrets set, and service bindings OUTREACH/AVA/RACHEL. First briefing surfaced a 70% bounce-rate finding.
Live
LEAD_KV — Shared Memory — DONE 2026-05-25
Namespace id 1bd4fc627b8d4bdbac8216c0a2a12ea5 bound to Outreach (writes COLD on send), Rachel (escalates to WARM/INTERESTED on reply), Faye (bumps emailsSent on follow-up), Morgan (reads funnel snapshot in briefings). Status only escalates — never demotes. Inline upsertLead helper mirrored across all 4 workers.
Live · Level 3
Faye — Follow-Up Agent — DONE 2026-05-25
Worker automatedrcm-faye live with FAYE_KV (f53f3df1…85a2e) + LEAD_KV bindings, cron 0 12 * * * (daily 8 AM ET), reads LEAD_KV for WARM/INTERESTED stale ≥3 days (caps 4 follow-ups/lead, 25 sends/run), drafts via Claude Haiku, sends via Gmail (reusing Rachel's OAuth), emails Laureen a per-lead summary.
Live
Sentinel — Security & Health Monitor — DONE 2026-05-25
Worker automatedrcm-sentinel live with SENTINEL_KV + service bindings to all 5 workers. Daily 9 AM ET cron pings every worker; Sundays branch into weekly credential audit (Anthropic, Gmail OAuth, Twilio, LinkedIn). Emails Laureen on any failure; weekly audit always emails. First run: 5/5 workers healthy, 12-20 ms latency.
Live
All 6 agents have real /chat (Level 4 complete) — DONE 2026-05-25
Outreach, Ava, Rachel, Morgan, Faye, and Sentinel all expose POST /chat → returns Claude-Haiku response grounded in each agent's own KV state. CORS preflight on all 6. Dashboard sidebar talks to all of them as real two-way conversation. The era of canned chat responses is over.
Live · Level 4
Morgan as Hub Agent — DONE 2026-05-25 (expanded to 7 siblings)
Every Morgan /chat fans out to all 7 sibling agents (Outreach, Ava, Rachel, Faye, Sentinel, Greta, Finn) in parallel via service bindings, then synthesizes one unified answer with Claude Haiku. ~8 Claude calls per chat (~$0.007), ~10-18 s latency. Failure on one sibling doesn't block the response.
Live · Hub · 7-way
Rachel LLC/EIN Email Watcher — DONE 2026-05-25
Rachel now runs a second Gmail query alongside her normal reply scan that explicitly bypasses the noreply filter — catches legitimate emails from noreply.efile@irs.gov, efile@sunbiz.org, and Florida Division of Corporations + any email with LLC/EIN/Sunbiz subject keywords. On match: instant Twilio SMS to Laureen's phone + priority email (X-Priority: 1) + LEAD_KV flag. Deduplicates against llc_alert_ids in REPLY_KV so the same email never alerts twice.
Live
Greta — Grants Agent — DONE 2026-05-25
Worker automatedrcm-grants live with GRANTS_KV (39215ba1…e130) + LEAD_KV bindings. Cron 0 11 * * 1 (Monday 7 AM ET). Curated list of 20 verified no-fee programs hardcoded (NIH SBIR/STTR, NSF SBIR, ARPA-H, SBA Growth Accelerator, Amber Grant, Cartier Women's Initiative, Tory Burch Fellows, Hello Alice, Florida WBC, etc.). Filters by current LLC/EIN/SAM.gov status + 90-day deadline window. Claude Haiku ranks survivors. Weekly email. Never hallucinates programs.
Live
Finn — Funding Agent (no loans) — DONE 2026-05-25
Worker automatedrcm-finn live with FUNDING_KV (80088d90…e449) + LEAD_KV bindings. Cron 0 11 * * 1 (Monday 7 AM ET). 20 curated non-loan programs: equity-free credits (NVIDIA, Microsoft, Google, AWS), equity accelerators (YC, Techstars, MedTech Innovator, Cedars-Sinai, AlchemistX), pitch competitions (Pioneer, Visa Everywhere), women-focused VC (Backstage Capital, Female Founders Fund, Pipeline Angels), and government R&D contracts. NO LOANS enforced at curated-list and Claude-prompt level.
Live
SMS Bridge — Outbound DONE · Inbound NEXT
Outbound (shipped 2026-05-25): Rachel fires a Twilio SMS to your cell on every INTERESTED classification, in parallel with the email alert. /test-sms endpoint available for synthetic testing.
Inbound (still planned): Twilio receives your text → dispatch worker routes to the right agent's /chat → agent responds via Claude with full context → Twilio texts you back. You text "Ava what did you post today?" and she tells you.
Half Done
Inbound (still planned): Twilio receives your text → dispatch worker routes to the right agent's /chat → agent responds via Claude with full context → Twilio texts you back. You text "Ava what did you post today?" and she tells you.
Vera — Eligibility Verification Agent
When a practice onboards, Vera checks patient insurance eligibility before appointments using the payer's API (or screen-scraping Availity with Claude in Chrome as a fallback). No more manual eligibility checks.
Phase 2
Coda — AI Medical Coder
Reads clinical notes, suggests ICD-10 and CPT codes, flags missing documentation before claim submission. Runs as a Level 2 agent — doctor submits notes, Coda returns a coded encounter for review.
Phase 2
Dex — Denial Management Agent
Monitors EOBs for denied claims. Classifies denial reason. Auto-drafts appeal letters using the payer's own rules. Tracks appeal deadlines. This is where the biggest revenue recovery lives — denied claims are money left on the table.
Phase 2–3