Work from chat
Ask the agent in messaging apps such as Telegram, WhatsApp, Slack or Discord, or from a web console.
PAI-AGENT
Private AI Agent Platform
Agents that get work done inside your walls, with you in control.
Hand routine work to AI agents that run on your own infrastructure. Built on the open-source Hermes Agent or OpenClaw, they take requests from the messaging apps your team already uses, remember context, learn reusable skills and run scheduled tasks. Logic Sonata wraps them in a governed sandbox: private models, allow-listed tools, human approval for sensitive actions and a complete audit trail.
- Built on Hermes Agent or OpenClaw
- Private models, sandboxed tools
- Human approval and full audit trail
01The challenge
Agents are powerful. Ungoverned agents are a risk.
Open-source agents can read files, browse the web, run commands and send messages. That is exactly what makes them useful and exactly why they must not run with unchecked access to company systems or public AI services. The value comes from pairing capable agents with strict governance.
02Capabilities
What it does.
Memory and skills
Agents remember context across sessions and turn solved problems into reusable, readable skills.
Scheduled tasks
Daily reports, weekly summaries and recurring checks, described in plain language.
Sandboxed tools
Files, spreadsheets, browser and scripts run in an isolated environment with allow-listed access.
Human approval
Sending email, updating records or spending money waits for a named approver.
Full audit trail
Every plan, tool call, approval and result is recorded and reviewable.
03How it works
From question to answer, inside your walls.
Hermes Agent from Nous Research or OpenClaw, both open-source, self-hosted agent frameworks with memory, skills, scheduling and messaging integrations. Logic Sonata selects the framework per use case and adds the governed runtime.
- 01TriggerChat, schedule or event
- 02PlanPrivate model on the appliance
- 03Sandboxed toolsAllow-listed access only
- 04ApprovalHuman signs off
- 05Deliver and auditResult, log, new skill
- 01
Trigger
A message, a schedule or an event starts a task.
- 02
Plan
The agent breaks the task into steps using a private model on the appliance.
- 03
Act in the sandbox
Approved tools run in isolation with only the access that task needs.
- 04
Ask before it matters
Sensitive steps pause for human approval in chat or the console.
- 05
Deliver and learn
Results are delivered, logged and captured as reusable skills.
04Example skills
Work your agents can take on.
Skills are readable, reusable instructions the agent builds and improves. These are typical first skills for manufacturing and trading companies.
Every morning
Read yesterday’s production export, calculate output, efficiency and defects by line, and post a summary to the operations channel.
On request
Collect the buyer’s tech pack, past quotations and material prices into a first-draft RFQ response for review.
Every Friday
List purchase orders overdue by more than three days and draft supplier reminders for approval.
After meetings
Summarise the notes, extract decisions and owners, and post them back to the team.
05Use cases
Where it pays off first.
Daily production and sales reports
Pull exports from ERP or MES, calculate KPIs and deliver a summary before the morning meeting.
RFQ and tender preparation
Collect specifications, past quotes and requirements into a first draft for review.
Supplier follow-ups
Track open purchase orders and draft chasers for approval.
Inbox and document triage
Classify incoming requests, extract key data and route them to the right team.
Meeting notes and actions
Summaries, decisions and owners, posted back to the team channel.
Market and competitor watch
Research through approved sources, summarised weekly.
06Security and governance
Controls built in, not bolted on.
Every Logic Sonata system runs on hardware you control, with access, audit and data boundaries designed in from the start.
- Agents run on private models, never public AI services
- Isolated sandbox with allow-listed tools and network access
- Credentials held in a vault, never exposed to the model
- Approval gates for external and irreversible actions
- Complete, reviewable audit trail
- Instant pause and kill switch
07FAQ
Questions about PAI-AGENT.
Hermes Agent or OpenClaw: which do we get?
Both are capable open-source agents. We recommend one based on your channels, tasks and security requirements, and deploy it inside the same governed runtime.
Can an agent act without permission?
Only within the limits you set. Read-only and internal steps can run automatically; anything external or irreversible waits for approval.
Which systems can agents connect to?
Files, spreadsheets, email, messaging apps and business systems with an API or export. Each connection is allow-listed and scoped.
08More from Logic Sonata
Other private AI systems.
Private Knowledge Assistant
A governed assistant that answers questions from your SOPs, specifications, policies and reports. Every answer cites its source and every user sees only what their role allows.
Private Vision Intelligence
Real-time analysis of live camera feeds by a vision-language model on your appliance. Ask what is happening on a line, at a dock or at an entrance, without video leaving your premises.
Private Coding Assistant
Code completion, repo-aware chat, review and agentic coding in the tools your developers already use, powered by open coding models on your own hardware.
Private Image Generation Studio
A ready-to-use image generation studio built on ComfyUI, preinstalled with curated models and templates, so designers create prints, concepts and campaign visuals without their work leaving the building.
Get started
See PAI-AGENT on your own data.
Book a 30-minute consultation. We will show the system working and scope a pilot around your use case.