An isolated AI runtime built for real business workflows
Omni PAWS is an opinionated stack — LangGraph Agent Core, secure MCP bindings, and a private VPS for every client. Here's how the pieces fit together.
LangGraph Agent Core — the brain of every deployment
Each client gets a LangGraph-powered agent that listens to your channels, follows your SOPs, and takes action through verified bindings — not public APIs.
Conversational intake
Greets leads on WhatsApp & Instagram in their language, asks the right qualifying questions, and never lets a hot lead go cold.
Tool-using agent
Reads, writes, and triggers actions via MCP — logging to Plane, creating invoices in Odoo, posting to your CRM.
Memory & SOP
Your tone, escalation rules, and guardrails are encoded in SOUL.md and MEMORY.md — and version-controlled like code.
MCP bindings to the tools your team already uses
We don't ask you to migrate. We connect to what you have, through encrypted channels you control.
n8n Free
Visual workflow automation. The agent orchestrates multi-step flows across your tools without writing glue code.
Supabase Free
Open-source Postgres with auth and storage. The agent reads and writes structured data through a single, verifiable layer.
Cal.com Free
Open-source scheduling. The agent books, reschedules, and confirms meetings on your team's calendar.
Slack
Team messaging. The agent posts updates, answers questions, and escalates to humans inside your channels.
Notion
Docs & knowledge base. The agent keeps runbooks, wikis, and SOPs in sync as processes change.
Stripe
Payments & billing. The agent generates invoices, checks payment status, and reconciles transactions.
WhatsApp & Meta API
Direct, verified bindings to WhatsApp Business and Instagram DMs via the official Meta APIs — no scraping, no copy-paste.
Plane
Project & task management. The agent logs status, assigns tasks, and syncs milestones — without anyone touching Plane manually.
Odoo ERP
Accounting, invoicing, and reporting — read-only by default, with explicit write authority scoped to high-trust actions.
Your SOPs become the agent's source of truth
We sit with your team, capture the way work really flows, and translate it into a SOUL.md + MEMORY.md pair that the agent loads on every conversation.
SOUL.md — tone & persona
Brand voice, escalation rules, taboo topics, language preferences — encoded so the agent never sounds generic.
MEMORY.md — operations knowledge
Your pricing, inventory, FAQs, team org chart, and approval matrix — refreshed as your business changes.
Continuous learning
Human-in-the-loop overrides feed back into MEMORY.md so the agent gets sharper every week.
The technical stack
Built on boring, dependable infrastructure so you can focus on outcomes — not vendor lock-in.
LangGraph Agent Core
Open-source agent runtime with deterministic tool routing and replayable audit trails.
Model Context Protocol (MCP)
Standardised bindings to WhatsApp, Meta, Plane, Odoo and your internal APIs — no scraping, no copy-paste.
DeepInfra LLMs
Hosted inference with predictable per-token pricing — open-source models, no data retention, no training.
Single-tenant VPS
One dedicated Linux server per client, in the region you choose, with full disk encryption and SSH-only access.
We run Omni PAWS on our own operations
Before selling AI to clients, we deployed our agent framework internally. Here's the before-and-after.
| Metric / Workflow | Before Omni PAWS | After Omni PAWS |
|---|---|---|
| Lead Response Time | 2–4 hrs manual delay | <30s instant WhatsApp/IG qualification |
| Project Tracking | Manual task logging in Plane | Automated logging & status sync via MCP |
| Operational Overhead | 15+ hrs/week on admin | 90% reduction in manual entry |
| System Uptime & Privacy | Shared SaaS dependencies | 100% isolated, single-tenant |
Five phases, one predictable rollout
No big-bang launches. We de-risk each step with a sandboxed UAT before anything touches production.
Discovery
30-min call. Map your top 3 manual workflows and rank automation candidates by ROI.
Info Gathering
We shadow your team, capture SOPs, and codify SOUL.md + MEMORY.md together with you.
Provisioning
Your dedicated VPS is stood up, MCP bindings configured, and LLM keys vaulted.
Sandboxed UAT
48-hour trial in an isolated environment. Your team pokes at it; we fix what breaks.
Deployment
Promote to production, wire monitoring, and hand over the runbook. You're live.
From a single inbox to a private AI workforce
Most automation vendors sell bots. We sell a workforce — isolated, governed, and trained on your operations.
Replace 2–3 FTEs of admin work
Lead qualification, project logging, invoicing, and status updates — handled before your morning coffee.
Keep your customer trust
Private VPS, zero leakage, no shared SaaS. Your customers never know an AI is in the loop — except by how fast you reply.
Predictable monthly cost
Flat per-tier pricing with usage telemetry. No surprise LLM bills, no per-seat creep.
Curious what your private AI workforce would look like?
Book a discovery call. We'll draw the architecture for your top 3 workflows and quote a fixed-scope UAT.