Seed Round Closed: $4.5M Led by Vertex Ventures · Total funding is now $6.7M. Investor and press coverage is live on Reuters, CRN, and Channel Dive.
FlamingoASCII Edge · San Antonio 2026
A keynote for MSP owners
The Highly Profitable and Self-Running MSP
Two bills, two levers, one plan.
9:14
Why I am up here
I did not start in MSP tooling. I watched MSPs get squeezed from the security side.
Where I startedVicariusThin marginsat our MSP customersVulnerability remediation
→
What owners were doingOpenMSPopen-source tools mapped members in its Slack
→
TodayFlamingolive deploymentsAI agents at the core
Everything in this talk, I learned from MSPs first.
I built Vicarius. Our customers were MSPs on thin margins.
Owners were swapping paid tools for open source on their own. Mapping it became OpenMSP.
Then we raised money and built AI agents into the core.
The real numbers
Two bills decide your margin: the tools you rent and the people you pay.
Bill 1 · The tools - 29% of revenueBill 2 · The people - another 24%Both bills are about to move.
↓ open source
The tools take 29% of revenue - ~$730K on a $2.5M shop.
The repetitive work takes another 24% - ~$600K. Together they leave you a 10-15% net margin.
One lever for each bill. That is the rest of this talk.
The MSP math problem
More tickets used to mean more techs. That equation stopped working.
The old model: more clients, more tickets, more techs.
Productivity is stuck at under half of what the best MSPs run.
You cannot hire your way out of a hiring shortage.
The margin is already breaking.
Who is coming
Investors are buying MSPs to rebuild them around AI.
The squeeze
Vendors raise your floor, wages raise your ceiling, and roll-ups are pulling your price down.
→
←
Three forces decide your margin. Only one of them used to be yours.
Vendors push your costs up from one side.
Wages push from the other side.
Your price was the one number you controlled. Now it is heading down. That is why the two levers are not optional.
Lever 1 · the tools bill
Every tool category you rent now has a production-grade open-source alternative.
Lever 1 · the rule
Whoever owns your telemetry captures your AI margin.
CopilotsPer userHelp one tech
Ticket agentsPer ticketClose the easy ones
Complex agentsPer tokenThe more they do, the more you pay
The 1 hour tool audit
- Cost per endpoint
- Overlap
- Utilization
- Exit terms
Copilots help one tech, and you pay per seat.
Ticket agents close the easy tickets, and you pay per ticket.
Complex agents do the real work on your data, inside their platform, and you pay per token.
Takeaway 1, this Monday: one spreadsheet, every tool, four columns.
Lever 2 · the people bill
Self-running is a ladder, not a switch. Most MSPs are on the first rung.
↑ most MSPs are hereAssistLevel 1 · assist
AISuggests the fix
HumanDoes every fix
AIDrafts the reply
HumanReads every alert
AISummarises the ticket
HumanWrites the notes
AIFinds the runbook
HumanCloses the ticket
SupervisedLevel 2 · supervised
AIDiagnoses the cause
HumanApproves each action
AIExecutes the fix
AIRestarts the service
AIPushes the patch
HumanReviews the change
AIClears the disk space
AIRotates the credentials
AIOpens the change record
HumanHandles the exceptions
AIWrites the ticket note
↑ the targetAutonomous, with guardrailsLevel 3 · governed autonomy
AIDetects before tickets
AIFixes and verifies
AIReports the result
HumanSets the guardrails
AICorrelates every tool
AIRolls back on failure
AIPatches the fleet overnight
AITunes the alert rules
AIWrites the postmortem
AICloses the ticket
HumanAudits exceptions
AIKeeps the audit trail
AILearns from every fix
AIEscalates what it cannot fix
AIRuns the overnight checks
AIIsolates the infected host
HumanOwns the customer call
AIRestores from backup
AIChases the failed agent
AIRenews the certificates
Lever 2 · where to start
Start with the boring work: high volume, low risk, easy to undo.
The work, with and without an agent
Routine Work
Without AI
With AI agents
ROI
Time Saved
ReversibleAuto-actRestarts, cleanups, patches
IrreversibleApprove firstDeletes, access, firewall
Real routine tasks, timed and priced with and without an agent.
If you can undo it, the agent just does it.
If you cannot, it asks first. Takeaway 2: tag 30 days of tickets, automate your top 5.
The catch
tools with copilots is still consoles. AI needs one stream.
different products
generating data
→
One stream · IT + security events
The dataReadable silo by siloEvery tool, its own console
The actionsLocked vendor by vendorA copilot that cannot act is a search box
Where does your routine work actually come from?
Every product you rent writes its own data, in its own console.
Help desk and security work is one stream of the same routine work.
Each copilot only sees its own tool.
Both levers need one data layer.
How it runs itself · OpenFrame
One data layer finds the problem nobody reported, fixes it, and asks your phone only when it matters.
tools, every log - one data layer
OpenFrame Core
Open Source
Raw logs
→
A user reports an issue
FaeCommercial
Nobody reports it - ML flags the anomaly
Live Analyzer
Commercial
Anomalies
→
Fixes it end to end, with guardrails
Auto Act
Commercial
→
Resolved - across every client
AI labor
Fixes done
Monday morning
Ninety days, four moves, no new vendor required.
Week 1Day 30Day 60Day 90
- Know your two numbers
- Endpoints per tech
- Gross margin per client
- Cut one tool, tag tickets
- Drop one overlapping tool
- Tag 30 days of tickets
- Automate your top 5
- Runbooks for your top 5
- Suggest first, then act
- Reprice and reinvest
- Reprice on the new margin
- Freed hours into security
The self-running MSP
Own your stack. Let agents do the routine. Run it from the bar.
raised · seed led by Vertex Ventures
live OpenFrame deployments
SOC 2 Type II
Coming soon
Backed by
Run your own numbersScan for the calculator
9:14
01 / 15↑ ↓ ← → or scroll to move