AJERIS
TourBlogDocsLog inText us
Log inText us
← AI agents

Self-hosted vs hosted AI agent: the honest math

Open-source AI agents are free as software, not as operations. Realistic time, cost, and reliability math for the self-host vs hosted decision.

Ajeris team·May 6, 2026·5 min read
Two parallel paths in soft natural light, the visual of the choice between operating the stack yourself and having it operated for you.

The dominant marketing narrative around open-source AI agents in 2026 is "free as in beer." OpenClaw is free; Hermes Agent is free. The downloads are free; the source code is free; the Discord is free. The implication is that the math against hosted SaaS is obvious: zero is less than non-zero.

The math is not actually that simple. The software is free; the operations are not. This post is the honest version.

★The thesis

The self-host vs hosted decision is a TCO decision, not a license decision. Self-hosting saves on subscription cost and pays in operational time. Hosted saves on operational time and pays in subscription cost. The right pick depends on your hourly rate and your tolerance for being on call for your own infrastructure.

What "free" actually costs

Self-hosting OpenClaw or Hermes Agent at production quality involves real recurring costs:

  • VPS or self-hosted server. USD 10-50/month at small scale, up to USD 200+ at scale. The agent has to run somewhere.
  • Model API costs. Pass-through to OpenAI, Anthropic, etc. Highly variable. Light personal use: USD 5-20/month. Heavy use: USD 50-150/month. Business use: hundreds.
  • Channel fees. Twilio for SMS/voice; WhatsApp Business; etc. Channel-by-channel pricing. Minimum USD 5-15/month for a personal-use phone number plus per-message costs.
  • Storage and bandwidth. Memory grows; conversation logs grow; image attachments grow. Negligible at first; meaningful over years.

Total recurring software cost: USD 30-300+/month, depending on usage. Not zero.

Plus the time costs:

  • Setup. 4-12 hours typical for non-engineers (most of whom give up). 1-3 hours for engineers.
  • Configuration tuning. A few hours over the first week as you discover what is broken.
  • Ongoing operations. 0-5 hours/month, mostly absorbing breaking changes from the project, debugging when something stops working, patching security issues.
  • Incident response. When something genuinely breaks (a model API change, a channel rotation, a skill regression), the time cost is what it is. 0-20 hours per incident.

For an engineer billing USD 100/hour, the operational time alone runs USD 100-400/month at typical incident frequency.

Realistic TCO: USD 130-700/month for a serious self-hosted personal agent.

What hosted actually costs

Ajeris is free to start and then usage-based. For most personal users, monthly usage works out to:

  • Light personal use (a few interactions per day): USD 0-25/month.
  • Heavy personal use (dozens per day, plus meeting recap, plus voice receptionist for business): USD 30-100/month.

Plus zero operational time. The 3am call is ours; the model API rotation is ours; the channel webhook fix is ours; the security patch is ours.

Realistic TCO: USD 0-100/month for a serious hosted personal agent. For most users, somewhere in the middle.

The TCO comparison, summarized

For a typical personal use case, the realistic monthly TCO comparison:

  • Self-hosted: USD 130-700/month (including operational time at engineer billing rates).
  • Hosted: USD 0-100/month.

The difference is mostly the operational time. If your operational time is essentially free (you genuinely enjoy operating the stack and you would have spent that time anyway on a hobby), self-hosted comes in cheaper at the bottom of the range. If your operational time is worth USD 50/hour or more, hosted is the cheaper pick.

For business use cases (voice receptionist, customer memory, lead capture), the math gets more lopsided. Self-hosting a business agent at production quality is hours-per-week of operational time forever. Hosted is set-it-and-forget-it.

What you give up by going hosted

Three things, all real:

  • Open source. If you want to read every line, fork the project, audit security, contribute back, hosted SaaS is the wrong shape.
  • Data residency. If your data has to live on your infrastructure for policy reasons, hosted is not appropriate.
  • The operational ergonomic. Some engineers genuinely enjoy operating the stack. That is a feature.

If any apply, self-host.

What you give up by self-hosting

Three things, also real:

  • Time. The biggest cost is hours. The hours that go to operations are hours that do not go to the work the agent is supposed to free you for.
  • Reliability ceiling. A hosted product runs 24/7 with on-call rotations and SLOs. A self-hosted product runs as long as you are awake and your VPS is up.
  • Updates and patches. Hosted products absorb upstream changes silently. Self-hosted products need you to absorb each one manually.

If those are not deal-breakers for you, self-host. If they are, hosted is the more honest pick.

A workbench with tools laid out neatly, the visual of the operational discipline self-hosting requires.
Self-hosting: the workbench is yours. Hosted: the workbench is ours.

When self-hosting is the right call

Concrete scenarios:

  • You are an engineer with operational competence and the time/inclination to use it.
  • You have data-residency requirements that make hosted SaaS unacceptable.
  • Open-source license matters to you for principled reasons.
  • You have a specific feature need that only the OpenClaw marketplace covers.
  • You enjoy the work as a hobby; the operational time is a feature, not a cost.

When hosted is the right call

Concrete scenarios:

  • You are not an engineer, or you are an engineer who wants your evenings back.
  • You want the agent's value unlocked from minute one without configuration.
  • You want a phone number, voice receptionist, or business agent on top of personal use.
  • The non-technical users in your life need to reach the agent.
  • You would rather pay USD 50/month than spend 5 hours debugging.
Self-hosted (OpenClaw or Hermes Agent)
  • Software free; ops time real.
  • Realistic TCO USD 130-700/month including ops at engineer rates.
  • You take the 3am call.
  • Open-source license; data residency yours.
  • Best for engineers with the time and inclination.
Hosted (Ajeris)
  • Free to start. Then usage-based.
  • Realistic TCO USD 0-100/month for personal use.
  • We take the 3am call.
  • Closed-source SaaS; data hosted by us.
  • Best for non-engineers and engineers who want the evenings back.
4-12 hrsTypical OpenClaw setup time
5 minAjeris setup time
~10xTCO difference for engineers at typical billing rates

Frequently asked

Are these numbers conservative or aggressive?+
The self-hosted side is at the conservative end of what we have observed. Realistic operational time for engineers running OpenClaw or Hermes Agent at production quality is often higher than the 0-5 hours/month we cited. Some users report dramatically more during incidents.
What if I have free hosting from a credit?+
Cloud credits do not cover operational time, model API costs, or channel fees. The math improves but does not flip.
Open source AND hosted?+
A few products try this. Most end up making compromises in one direction or the other. We are closed-source SaaS for engineering simplicity; that is a deliberate choice we are not trying to defend on principle.
Can I switch from self-hosted to hosted later?+
Yes. Most users who switch do so when they realize the operational time has been more expensive than they thought. The friction is low: text the agent and walk through the setup wizard.
What about a hosted version of OpenClaw?+
Some companies offer hosted OpenClaw deployments. The hosted-vs-Ajeris comparison still applies; the differences are in the contract (skill marketplace vs vetted catalog, broad permissions vs layered approval, etc.).

Try the math on your specific case

Free to start. Text +1 (470) 821 4265. The honest test is comparing the operational time you currently spend on OpenClaw or Hermes Agent against the usage-based bill on Ajeris over a typical month.

For most users, the math works out cleanly in one direction or the other.

More stories

A phone on a cafe table, top-down. The case for the agent that lives where the conversation already is.
AI agents

Why texting wins: the case for messaging-first agents

Chat windows, smart speakers, and dedicated apps each have a thesis for the AI agent. The thesis that holds up daily is the one that meets people in their messages.

An iPhone on a cafe table with a serif headline reading 'An AI agent that lives in your text messages.'
AI agents

What is a personal AI agent?

A personal AI agent is one assistant that lives in your text messages and actually does the work across every tool you already use. Here is what that means.

Try Ajeris.

Text +1 (470) 821 4265. Your agent replies in seconds. Free.

Text on WhatsApp
Connects to

Every tool you already use. Your agent does the rest.

WhatsApp
Gmail
Calendar
Slack
Notion
GitHub
Instagram
Spotify
Uber
DoorDash
Apple Music
Telegram
WhatsApp
Gmail
Calendar
Slack
Notion
GitHub
Instagram
Spotify
Uber
DoorDash
Apple Music
Telegram
Shopify
QuickBooks
Xero
Square
HubSpot
Salesforce
Stripe
Lyft
Alexa
Google LSA
Hue
Claude
Shopify
QuickBooks
Xero
Square
HubSpot
Salesforce
Stripe
Lyft
Alexa
Google LSA
Hue
Claude

Message to start

+1 470 821 4265

Message the number above on WhatsApp or by SMS. Your assistant replies within seconds and sets itself up in the conversation. Or sign up by web form to be added to our list.

Welcome message. “Hi. I’m Ajeris. I’ll be your personal agent. What should I call you? Reply STOP to unsubscribe, HELP for help.”

STOP to unsubscribe. HELP for support.

Message and data rates may apply. Carriers are not liable for delayed or undelivered messages. Message frequency varies based on your interaction with the agent.

Questions: hi@ajeris.com

Your assistant is already in your phone. You just need to text it.

The personal assistant you can actually get today. No app, no waitlist.

WhatsApphi@ajeris.com
Personal
  • Channels
  • Morning brief
  • Meetings
  • Research and build
  • Laptop on a leash
  • Everyday
  • Memory
Platform
  • Connectors
  • Platform
Developers
  • Docs
  • Quickstart
  • Platform overview
  • Memory model
  • Skills
  • Scheduled tasks
  • MCP server
  • Claude Code
  • Alexa+
Resources
  • Blog
  • Compare
  • Glossary
  • Live webinar
  • Creator Program
Trust
  • Privacy
  • Security
  • Terms
  • Data retention
  • Access controls
  • Vulnerability management
  • Opt in
  • Disconnect

Your personal agent, by text.

Try it free·Ajeris for business·docs.ajeris.com·© 2026 Ajeris