Private AI on your stack.
Live in 48 hours.
I deploy a personal AI assistant onto your WhatsApp, Telegram, and Slack — running on your hardware, owned by you, set up by hand. No SaaS, no data sharing, no vendor lock-in.
Yes, it is possible. You can run a real AI assistant on your own server — answering customers on WhatsApp, Telegram, or Slack — without sending conversations to OpenAI or paying per-message fees. The catch is the setup: ~40 hours of OpenClaw, Claude Code, channel pairing, agent bootstrap, hooks, and integrations most teams cannot spare. Prayogai does it for you in 48 hours and hands you the keys.
You don't want another chatbot.
You want an AI that knows your business, talks to your customers on the channels they actually use, and doesn't ship every conversation to OpenAI's training set. The open-source stack to do this exists — OpenClaw, Claude Code, local LLMs, the right glue. The hard part isn't picking tools. It's the setup.
WhatsApp pairing. LaunchAgent vs systemd. Agent bootstrap. Channel routing. Memory schemas. PreToolUse hooks. Per-sender session scope. Gateway auth, reverse-proxy headers, sandbox modes, OAuth flows, Telegram bot tokens, cron drift, voice transcription, TTS rate limits. Each one is a four-hour rabbit hole. There are forty of them.
You don't have those weekends. You have a business to run.
An assistant. Not a chat window.
System
- OpenClaw gateway, hardened, your hardware or your VPS
- Channel pairing across WhatsApp, Telegram, Slack, iMessage as needed
- Memory + vector store + FTS, encrypted at rest
- Per-sender session isolation, daily auto-rotation
- Reverse-proxy + auth token rotation, audit-clean by handoff
Agent
- Custom persona configured around your actual use case
- Workspace + bootstrap files (AGENTS, IDENTITY, USER, TOOLS)
- Tool allowlist + deny rules + approval hooks for sensitive actions
- Optional integrations: Calendar, Gmail, GitHub, Stripe, your CRM
- Voice in / voice out (Deepgram + ElevenLabs) if you want it
Handoff
- Operating runbook: how to restart, where logs live, what each cron does
- Health checks + status notifications to your phone
- Backup + restore procedure tested in front of you
- 30-day tuning window: I tune the prompts as you collect real usage
- Source-of-truth config in a private git repo you own
Ongoing (optional)
- Monthly: monitoring, updates, security patches
- Two scoped change requests per month included
- Slack/WhatsApp access to me for production issues
- Quarterly tuning review with usage analytics
- Cancel any month, no lock-in, you keep everything
Why own beats rent.
Most "AI for WhatsApp" products are SaaS — your data lives in the vendor's cloud, you pay per message or per minute, and you're stuck with their pricing forever. Self-hosted is the opposite shape. Here is how it compares for a typical small-business volume.
| Prayogai (self-hosted) | SaaS (ManyChat, Trillet, Newo, etc.) | |
|---|---|---|
| Where it runs | Your server, your VPS, your hardware | Vendor's cloud |
| Per-message cost | $0 — flat monthly | $0.005–0.20 per message or minute |
| Data ownership | Yours, on disk you control | Vendor's, exportable at best |
| Vendor lock-in | None — config in a git repo you own | High — proprietary platform |
| Custom integrations | Any tool, any API, full code access | Limited to vendor's marketplace |
| Setup | 48 hours, hand-tuned by Prayogai | Self-serve sign-up |
| Cost @ 1k msgs/mo | $349 flat | $99–300 + variable charges |
| Cost @ 10k msgs/mo | $349 flat | $500–2,000+ |
| Best for | Privacy-aware businesses · agencies · operators who treat customer data as theirs | Teams who don't mind SaaS lock-in |
Three sizes. Pick the one that fits.
Tier 1
- 1-day deployment to your Mac mini or VPS
- WhatsApp + Telegram channels, paired and tested
- One agent, your persona, your use case
- 1 hour of memory ingestion (your docs / prior chats)
- 30-day tuning included · 2 small changes / month after
Tier 2
- Everything in T1, on a dedicated VPS I operate
- 2–3 specialized agents (e.g. receptionist · outreach · ops)
- Calendar, Gmail, Stripe integrations as needed
- Approval workflows for sensitive actions
- Weekly tuning calls in month 1 · 4 changes / month after
Tier 3
- Multi-tenant deployment with profile isolation per client
- Branded dashboard for your team
- You sell to your clients · I run the backend
- Per-vertical agent templates (clinic, salon, contractor, &c.)
- Volume discount past 25 end-clients
Be honest about whether this is for you.
This is for you if
- You handle real volumes of inbound on WhatsApp / Telegram
- You won't put customer data into a third-party SaaS
- You can name three workflows you'd automate tomorrow if you could
- You'd rather pay once for a system you own than $99/mo forever
- You'd rather not become an OpenClaw expert yourself
This is not for you if
- You want a magic box with no operator on your end
- You expect AGI; this is a careful AI assistant, not a miracle
- You'll send fewer than 100 messages a month — overkill
- You need it deployed in a country I can't currently reach
- You want to underpay good engineering. This is hand-built work.
From call to live in seven days.
Discovery call
We map what you actually need: which channels, which workflows, which tools you already use. I tell you which tier fits, or that you don't need me at all.
Written scope
You get a one-page scope with specifics: agents, integrations, hosting target, success criteria. Approve it or push back. Nothing built until we agree.
Build & pair
I deploy on your hardware or VPS. Channels paired. Agent persona configured. Memory schema set. Cron and hooks wired. Backup tested.
Live handoff
Screen-share. I show you every moving part: how to restart, how to read logs, where to change prompts. You drive while I watch. We restart it once together.
Tuning window
You collect real usage. I refine prompts, add allowlist patterns, fix edge cases. Daily Slack/WhatsApp access to me. Most of the value happens here.
I run my own. I'll run yours like I run mine.
Prayog · प्रयोग · Sanskrit for "use, application, deployment"
Solo operator. I built this stack for myself first — proactive AI on WhatsApp running on a hardened OpenClaw gateway, Gemini routing, Claude for the heavy work, Brave for search, ElevenLabs for voice, all of it self-hosted. Friends wanted theirs. Now it's a service.
No team. No outsourcing. No project managers. You talk to me, I build it, I'm the one debugging at 2am if something breaks. Capacity is finite — I take five new clients a month, no more.
Honest answers to the obvious questions.
What is a self-hosted AI assistant?
A self-hosted AI assistant runs on hardware you control — your Mac mini, your VPS, or your colocated server — instead of a vendor's cloud. The agent connects to your WhatsApp, Telegram, or Slack via channel pairing. Conversations and customer data never leave your infrastructure unless you explicitly route them out (for example, when calling an LLM API).
The trade-off is ownership and privacy in exchange for setup work and operational discipline. Prayogai handles the setup work. You keep the ownership.
How much does it cost to deploy a self-hosted AI assistant on WhatsApp?
Prayogai's Personal AI tier is $349 per month, with the first month at $1,800 to cover hand-tuned onboarding (about 15 hours of senior engineering work). Business AI is $799/mo ($4,000 first month). Agency white-label is $1,799/mo plus $149 per end-client ($7,500 first month).
All tiers are flat-rate. No per-message fees, no per-minute charges, no surprise overages. A SaaS competitor charging $0.10/message costs you $1,000 at 10,000 monthly messages; Prayogai stays at $349.
How is Prayogai different from ManyChat, Trillet, Botpress, or other AI receptionists?
ManyChat, Trillet, Newo, Botpress, and similar platforms are SaaS — your data lives in their cloud, you pay per message or per minute, and you depend on their pricing and policies. The day they raise prices five-fold, your business notices.
Prayogai is the opposite shape. The AI runs on your hardware. You own the data. You pay a flat monthly fee with no per-message billing. There is no vendor lock-in — your config lives in a private git repo you control, and the work is fully transferable to any developer.
Why not just use ChatGPT or Claude.ai for my business?
ChatGPT and Claude.ai are excellent in a browser tab, but they don't sit on your WhatsApp, don't have memory across customer conversations, can't run cron jobs, can't act on webhooks, and ship every word through the vendor's infrastructure. They're products for you to use yourself, not your business's AI assistant.
What if WhatsApp blocks the bot?
Personal-account WhatsApp pairing tolerates AI-assisted use within reasonable rate limits — Prayogai configures conservative pacing, message-rate limits, and clear opt-in patterns. Past ~5,000 monthly messages or any client-facing volume, we migrate the deployment to WhatsApp Business API ($50–$500/mo platform fees depending on volume), which has full sanction for automation.
A fallback channel (Telegram or SMS) is included so a single-channel ban never takes the assistant fully offline.
What if I want changes after handoff?
T1 includes 2 small changes per month. T2 includes 4. T3 includes unlimited small changes for the agency stack. A "small change" is anything that takes me under an hour: tweak a prompt, add an allowlist pattern, wire one new contact, adjust a cron schedule.
Need a bigger redesign — a new agent, a new integration, a new workflow? $200/hr against an agreed scope, or roll into a higher tier and get it included.
Can I cancel? What happens to my data?
Cancel any month with no penalty. You keep the server, the configurations, the agents, and the memory store — Prayogai does not hold your data hostage.
Your config also lives in a private git repository under your control, so the work is fully transferable to another developer or in-house team. The only thing you lose by canceling is Prayogai's pager rotation.
If you cancel within 30 days of handoff and you're genuinely unhappy with the work, you get a refund minus infrastructure costs incurred.
How long does setup take?
From a kicked-off scope to a live deployment is typically 48 hours. The full process from first call to fully tuned is one week: discovery call (day 0, free, 30 min), written scope (day 1), build and pair (days 2–4), live handoff (day 5, 90-min walkthrough), tuning window (days 6–35).
How do you charge from outside the US?
Wise Business, direct bank wire, or Stripe (via a US LLC) — your pick. For US/EU clients, Stripe is most painless. For agencies on retainer, monthly invoicing via Wise. Crypto accepted for clients in regions where banking is friction. Ask.
The first call is free. Three slots open at 30% off.
Email me to book → hello@prayogai.comReply within 24h · No sales call energy · I'll tell you if you don't need me