A rigorous case for building the chiropractic vertical agent product on the AscendOps playbook, then beating it. Verified repo intel, the core product mechanic, a concrete rate card grounded in real clinic economics, the honest risks, and what to do tonight. Prepared for a real business decision.
Round-1 intel and the prior playbook both recorded the AscendOps repo as private, needing a $99/mo Skool membership to read. That is now disproven. The repo is live, public, MIT-licensed, and readable with no gate:
https://github.com/noogalabs/ascendops HTTP 200 · "private": false · MIT · TypeScript
noogalabs (created 2026-06-02, last pushed 2026-06-16), which is why the org-level API 404s and confused earlier checks. The repo itself resolves cleanly.# cortextOS. It is a near-verbatim fork of the same harness we run. Installer points at noogalabs/ascendops-install.SKOOL-INSTALL.md (the full zero-to-fleet install David charges to teach), RENEWALS-AGENT-DOSSIER.md (a complete community-contributed decision-tree agent), the community/ catalog and agent templates, the Telegram permission/crash/loop hooks in bus/, and templates/ for maintenance-director, leasing-coordinator, and renewals-coordinator.Strategic consequence: we do not need the Skool Level-2 gate to study his architecture. We can opensrc + graphify the public fork directly, lift the install playbook and the contributed-agent dossier structure, and skip the membership. The $99/mo is now optional intel, not a blocker.
David's model is sound and proven on his own 420-door operation, which is the strongest possible product-market-fit signal: the builder is the customer. The question is not whether the model works. It is whether we are positioned to run the chiropractic version better than anyone else who reads his public repo. We are, on five axes David had to build from nothing.
We run a 10-agent fleet on cortextOS daily: bus, crons, dashboard, memory, KB, browser hands. David is a student of this same harness. We operate it; he is still hardening his fork.
David spent months interviewing and encoding property-management judgment into decision trees. Zach is a chiropractor. The clinical and front-desk decision trees can be authored first-hand, faster and more authentically than any competitor can reverse-engineer.
David built a PM audience from zero. The AgenticChiro buyer is the TTC and Pediatric Experience audience Zach already reaches. We are not finding a market. We own one. The TTC Buffalo workshop (Jul 18-19) is a natural first stage.
Van Every is a real running clinic where the fleet already does intake-adjacent work, insurance audits, recall, and content. Productize what already works there. Proof before pitch, not a slide deck.
ChiroHD, the dominant chiro EHR, is browser-only: no API, no webhook, no export. SKED and GHL are patchy. Whoever builds the browser-operator rail for ChiroHD owns the automation layer for every ChiroHD clinic in the country. First-mover on an un-API'd EHR is a durable position, and we already run this exact pattern.
Clinics that buy contribute their own agents (pediatric intake here, auto-SOAP there) back to the repo, exactly as a Foothills PM operator contributed the renewals agent that now ships to every AscendOps user. But ours is seeded by Zach's existing clinical corpus (VE KST YouTube library, condition pages, PXDocs) that no competitor has.
A forked, chiro-branded cortextOS template that ships encoded clinic-ops judgment plus browser hands for the no-API EHR stack, controlled from Telegram, gated by an approval matrix. The core mechanic, lifted and improved from David, is SOP-to-Skill translation.
Every serious chiropractic practice already has SOPs, in the DC's head, in the front-desk binder, in "how we do things here." The gap to an agent product is not technical. It is translation. The mechanic:
SKILL.md with a pushy trigger description and an explicit decision tree. Routing logic (new vs returning, KST vs Talsky Tonal, acute vs wellness, urgency tiers) is codified before runtime so the agent executes Zach's expertise, not live AI guesswork.Agent reasoning, decision-tree execution, task state, memory, queues, coordination. Stable CPU and RAM that never competes with a browser. This is the same brains-in-a-datacenter split detailed in the VPS Value Case: it solves our own fleet's RAM and swap crunch and becomes the reference appliance we sell. Dogfood equals demo.
Playwright operators with isolated Chrome profiles, queue-based: reserve a profile, execute against ChiroHD / SKED / GHL, return result, release. Loop: VPS decides, Mac executes, VPS logs. Patient data never leaves the clinic's own appliance, which is also the HIPAA selling point.
Clinic profile schema, the decision-tree skills, the browser operators, the VPS-brains plus Mac-hands queueing, and a zero-to-fleet install playbook modeled on David's SKOOL-INSTALL.md two-phase structure (bootstrap, then per-agent onboarding interview).
| AscendOps (PM) | AgenticChiro (chiro) | Status today |
|---|---|---|
| Orchestrator | Front-desk / chief-of-staff agent | Run it daily (Sage pattern) |
| Analyst | Insurance-audit / KPI agent | Running at VE (Hippocrates) |
| Maintenance Director | Patient-intake-triage agent | Encode from VE intake SOP |
| Leasing Coordinator | New-patient comms / recall agent | Encode from VE recall flow |
| Property Meld Playwright (no API) | ChiroHD / SKED Playwright (no API) | Already running this rail |
| Renewals-coordinator (community-contributed) | Auth-renewal / reactivation agent | First clinic contribution target |
Starter skill set, derived by analogy from the public PM skills:
new-patient-intake-triage appointment-reminder recall-outreach (inactive 30+ days) insurance-verification-request missed-appointment-recovery claim-status-check EOB-processing patient-balance-notice authorization-renewal-reminder birthday / reactivation campaign
These are confirmed in both the classroom curriculum and the public repo. Lift them directly; they are the parts that make the product teachable and trustable.
Never write a CLAUDE.md or a skill from a blank file. "Interview me one question at a time about how this works, then write the SKILL.md with a pushy trigger description." This is also exactly how a new clinic onboards: the agent interviews the DC into existence.
Every skill must answer: (1) when does it trigger, (2) the exact ordered steps, (3) rules and thresholds, (4) what does done look like with an example, (5) what must NEVER happen. Question 5 is the one humans forget and agents need most. In chiro, the nevers are sharp: never give clinical advice, never promise an outcome, never touch a claim without approval.
Agent always carries every skill's name and description (cheap), opens the full body only when triggered (medium), reads bundled reference files only when needed (free until used). This is why 50 clinic skills do not drown the agent. Keep bodies under ~500 lines; push reference into a references/ subfolder.
Reversible+low-stakes: do it silently. Reversible+high-stakes: do it, log it, tell after. Irreversible+low-stakes: do it within thresholds. Irreversible+high-stakes: always escalate. For chiro, the bottom-right is everything patient-facing, insurance-facing, or PHI-touching. "Start tight, loosen with logged evidence."
Claude Code as planner, Codex as executor, told to "challenge each other on all plans and code like sisters born a year apart." David reports this produces some of the cleanest AI code he has seen. We already run this split; it ships as the build layer of the product.
David's own method for learning a repo: pull the real source with opensrc, turn it into a searchable knowledge graph with graphify, extract the patterns worth stealing. We point this at his public fork to lift the install playbook and the contributed-agent structure without paying.
RENEWALS-AGENT-DOSSIER.md shows a community member's agent encoded as explicit stages, each tagged "autonomous" vs "escalate to human," with named EHR fields and a 4-hour heartbeat that never sends external comms. That dossier is the AgenticChiro skill spec, with LeadSimple/AppFolio swapped for ChiroHD/SKED. Copy the shape, change the nouns.
David's pricing is partly un-anchored. He is still publicly price-discovering the done-for-you tier, asking his own community what to charge. That is the gap. Below is a concrete AgenticChiro rate card, grounded not in guesswork but in real VE patient economics: $200 new-patient exam, $50 consult, $2,500 average care plan. The ROI math writes itself: one recovered lapsed patient or recaptured no-show pays for months of managed service.
| Tier | Price | What it is | ROI anchor |
|---|---|---|---|
| Community | $0 | Free chiro-agent community, lead magnet ("Clinic AI Starter Kit": intake questionnaire, prompt pack, the interview method). Email capture and audience seeding. | Top of funnel |
| Repo / DIY | $149/mo | AgenticChiro repo access, weekly live builds, monthly deep-build, community-contributed skills. Priced above David's $99 because the chiro skills are seeded by a practicing DC. DIY friction is a designed upgrade trigger. | One recovered no-show/mo |
| Done-for-you install | $5,000 one-time | We provision the VPS brains, ship the Mac-mini hands appliance, wire ChiroHD / SKED / GHL, run the interview to encode the clinic's own SOPs, and hand over a working fleet. Matches David's anchor; defensible against his un-priced tier. | 2 recovered care plans |
| Managed | $500-$750/mo | Ongoing ops, agent and skill updates, monitoring, support, new-skill requests. $750 for multi-location or higher-volume clinics. | < one care plan/quarter |
| High-ticket (later) | $10K + $2,500/mo | Capped inner circle, multi-clinic groups. Mirror David's exploratory top tier, raise in steps as demand proves out. | Group / DSO economics |
Why these numbers hold. A clinic running $2,500 average care plans is already spending real labor dollars on manual EHR entry, recalls, and follow-ups, the exact tasks the product automates. The $5K install is two recovered care plans. The $500/mo managed fee is recouped by a single reactivated patient per quarter. Unlike David's PM math (which he is still figuring out publicly), we can state the payback in care-plan units a DC instantly understands.
An honest deliverable argues both sides before it recommends. Here is the sharpest bull case against the sharpest bear case, then the resolution.
Every bear point is real and survivable, and resolving them sharpens the product rather than killing it.
Verdict: build it, designed around the PHI posture from line one, sequenced to dual-purpose our own infra work, soft-launched to one clinic before any scale commitment.
Automating a live EHR with real patient data for clinics that are not ours is a different risk surface than VE's own marketing. Mitigation as feature: on-machine hands, no PHI leaving the clinic's appliance, encrypted brains, a BAA story. "Your data never leaves your clinic" is the headline selling point, designed in, not bolted on.
Browser automation against an EHR may bump terms. Mitigation: assistive framing, human-in-the-loop, rate-limited human-paced operation, and an export-and-email fallback so a hard block degrades the product instead of breaking it.
$5K installs and $500/mo managed are a services load with a hard concurrency ceiling. Mitigation: cap simultaneous installs, templatize hard, route the price-sensitive tail to the DIY repo, and reserve high-touch for the high-ticket tier only.
We learn David's public playbook to apply to a different vertical. Non-competitive in PM. Mitigation: keep it clean, build on the MIT-licensed harness, never resell his PM repo or his encoded PM rules, ship our own chiro skills.
opensrc + graphify the public AscendOps fork. Extract the install playbook, the contributed-agent dossier shape, and the hook scripts.