100% on-premise
Deploy inside your own data center or private cloud. Documents, embeddings, credentials, and every log line stay behind your firewall. Local LLM inference: no external AI API required.
On-premise AI Operations Platform
Your company's knowledge.
One message away.
Trevis sits on top of the systems you already run (shared folders, HR apps, ERP, spreadsheets) and turns the chat apps your team already uses into the work interface. Ask a question, get a cited answer from your source of record. Permission-aware. Fully audited. And your data never leaves your walls.
WhatsApp live
Slack coming soon
- 100% on-premise
- Zero new apps to learn
- Every answer cited
Trevis Assistant
online
How many annual-leave days do I have left this year?
You have 9 annual-leave days remaining for 2026.
Source: HRIS · leave-balance · checked just now
What's the Q2 sales total for the Jakarta branch?
Jakarta branch Q2 total: Rp 4.82 B (+12% vs Q1).
Source: ERP · sales-report-q2.xlsx
Orders API is throwing 500s in production. What's going on?
Found it: NullPointerException in payment-service v2.4.1, deployed 13:58.
Source: prod logs · error-monitor
Open a GitHub PR with the fix.
PR #482 opened for review:
github.com/acme/payment-service/pull/482
Action: github · pr-create
ROUTING TRACE
- role Finance · Viewer
- skill sales-report ✓
- perm use:sales-report ✓
- source ERP connector
- log audit #48213
0
bytes of data leave your infrastructure
100%
of answers grounded & cited
1
app your team needs: the one they already use
<1min
from question to sourced answer
The problem
Company knowledge is trapped in silos.
Company knowledge is trapped in silos.
AI chatbots make things up.
Answers live in shared drives, HR portals, ERP screens and spreadsheets, locked behind logins, formats, and tribal knowledge. Generic AI assistants either can't reach them, leak them to the cloud, or hallucinate. Trevis fixes all three.
Cloud AI = data exposure
Sending internal documents to third-party AI services is a compliance nightmare. Trevis runs entirely inside your perimeter.
Hallucinated answers
A confident wrong answer is worse than no answer. Trevis replies "not found" instead of guessing, and cites every fact it returns.
Yet another app
New tools die from non-adoption. Trevis meets your team where they already are: WhatsApp. Zero installs, zero training.
How it works
From question to cited answer, every step governed
Ask where you already chat
A team member sends a plain-language question (text or voice) in WhatsApp today, or Slack soon. No new app, no new habit.
Permission-aware routing
Trevis understands the intent and selects the right skill and integration, filtered by exactly what the asker is allowed to access.
Pull from source of record
The answer is retrieved live from your actual systems (folders, HR, ERP, Excel), never from a stale copy or a guess.
Reply with citation, log everything
The answer arrives with its source attached. The full trace (who asked, what ran, what was returned) lands in the audit log.
Why Trevis
Built for organizations that can't compromise
Every design decision starts from three constraints: data sovereignty, zero friction, and provable truth.
Chat-native: WhatsApp & beyond
No new app to roll out. WhatsApp is live today via self-hosted pairing or the official Business API, with Slack coming soon on the same channel layer.
Permission-aware, even reads
Two separate permission planes: console RBAC for admins, per-skill grants for WhatsApp users. Nobody sees what they shouldn't, not even by asking nicely.
Grounded & cited answers
Every reply carries its source. When the answer isn't in your systems, Trevis says "not found". It never fabricates.
Full audit trail
Every question, route, source, and answer is logged and searchable. Compliance reviews become a filter, not a forensic project.
From answers to actions
Shared folders, HR apps, ERP, spreadsheets, internal APIs, login-protected web apps: all connected through a registry with vault-stored credentials. Read skills answer questions today; action skills, like opening a GitHub PR from a production incident, are next on the roadmap, governed by the exact same permission, approval, and audit model. No rip-and-replace, no migration project.
Security & sovereignty
Your walls. Your rules.
Your walls. Your rules.
Your data, always.
Trevis was designed for regulated and privacy-conscious organizations from day one. It's not a cloud service with an "enterprise tier". It is infrastructure you own.
- Self-hosted everything Database, vector store, object storage, secrets vault, and LLM inference all run inside your network.
- Vault-secured credentials Integration credentials live in a dedicated secrets vault, never in config files or code.
- Allowlisted access only Channel users must be explicitly allowlisted (by phone number on WhatsApp), then granted skills one by one.
- Skill approval workflow New capabilities go through draft → review → approval before anyone can use them. Nothing ships silently.
Admin console
Total control, from one screen
A clean web console where IT owns the whole operation. No terminal required.
console.your-company.internal
0questions this week
0active skills
0integrations healthy
Try the agentChat simulator with a live routing trace: see role → skill → permission → source → log before going live.
Skills with AI-assisted draftingDescribe a capability; get a draft skill to review and approve.
Monitoring & auditEvery interaction searchable with filters: answers, denials, and not-founds alike.
Channel in-UIPair the WhatsApp channel by QR or connect the official Business API. No CLI.
The difference
Trevis vs. generic AI chatbots
| Trevis | Cloud AI chatbots | |
|---|---|---|
| Where your data lives | ✓ Inside your infrastructure | ✗ Third-party cloud |
| Wrong-answer behavior | ✓ Says "not found" | ✗ Hallucinates confidently |
| Source citations | ✓ On every answer | ✗ Rare or none |
| Access control | ✓ RBAC on every read | ✗ One bot sees everything |
| User adoption | ✓ WhatsApp today, Slack soon: already installed | ✗ New app, new logins, training |
| Audit trail | ✓ Complete & searchable | ✗ Opaque |
FAQ
Questions, answered.
Questions, answered.
With sources, naturally.
Does my company data leave our infrastructure?
No. Trevis is deployed on-premise or in your private cloud. Documents, embeddings, credentials, and logs all stay inside your own infrastructure, including LLM inference, which runs on local models.
Do employees need to install a new app?
No. Trevis works inside the chat apps your team already opens dozens of times a day: WhatsApp today, Slack soon. They send a message; they get a cited answer. That's the entire learning curve.
Which chat channels does Trevis support?
WhatsApp is supported today, via self-hosted pairing or the official Business API. Slack support is coming soon, and the channel layer is built to add more workplace chat platforms over time. Permissions, citations, and audit logging apply identically on every channel.
How does Trevis prevent hallucinations?
Every answer must be grounded to a source of record and is returned with a citation. If the information isn't found in your connected systems, Trevis replies "not found". It never invents an answer.
How granular is the permission model?
Very. Console admins get role-based permissions with direct grants on top. WhatsApp users are allowlisted by phone number and receive per-skill permissions. Each capability is granted explicitly, and every use is checked and logged.
Which systems can Trevis connect to?
Shared folders, HR applications, ERP systems, spreadsheets, internal APIs, and login-protected web apps. Connectors are registered in the console with credentials stored in a secrets vault.
Can Trevis take actions, or only answer questions?
Today Trevis focuses on read operations: answering questions from your source of record, with citations. Action skills are next on the roadmap: think identifying a production incident and opening a GitHub pull request with the fix, then sharing the PR link in chat for human review. Every action follows the same permission, approval, and audit model as reads. Nothing runs without an explicitly granted skill, and a human always reviews the result.
How long does deployment take?
Trevis ships as containerized services for your own hardware or private cloud. A typical pilot (one channel, a handful of integrations and skills) is live in days, not months.
Put your company's knowledge
one message away.
See Trevis answer real questions from your own systems, in a private, guided demo.
On-premise pilot · your infrastructure · your data never leaves