Integration & Control
AI your CTO and your compliance team can both approve.
We design and run governed AI agents for mid-sized companies: audit trail, access control, data masking and human approval built in from the first version.
Designed around the NIST AI RMF and ISO 42001 principles, and structured to support your own team's SOC 2 readiness work. Loucells Core is not SOC 2 certified.
The risks
What usually blocks AI inside a company.
Staff already use public AI tools
People paste customer and company data into chatbots to get work done. Until there's a governed option, that keeps happening.
No audit trail
An agent with no log and no access control is hard to defend in a security review or an audit.
Tools you can't inspect
Many AI products don't show what they do with your data or why they answered the way they did. That's hard to sign off on.
Control
We run the agent. You hold the keys.
Your phone number, calendar and CRM stay in your name. Revoke the keys and the agent stops right away. Conversations are stored encrypted and deleted when your retention period ends.
API keys · in your name
Owner controls
Try it · example panelAsk me before
Reminders respect quiet hours, 9 PM to 8 AM
Audit log
sha-256Rows are only ever added, never edited, and each one includes the hash of the row before it. We check the whole chain every day. Change a setting above and it shows up here.
Masked before the AI sees it
Messages with card or Social Security numbers are blocked, and emails and phone numbers are masked, before the model reads anything.
You approve the risky ones, by name
Refunds, discounts and quotes wait for an OK from you or the manager you choose. The log shows who approved each one.
A log that checks itself
Every action is chained to the one before it. We verify the whole chain every day and send you its fingerprint each Monday, so history can't be rewritten without it showing.
Every change to your agent is signed
Each version of the agent's instructions is saved with who approved it. You can see what changed and go back to any version.
It says it's an assistant
Your agent always introduces itself as a virtual assistant. If someone mentions an emergency or self-harm, it points them to 911 or 988 and alerts a person.
Deleted on schedule
Conversations are deleted automatically after 90 days, or the period you choose. The log keeps fingerprints, not messages.
Seven checks between a message and an action
- 1
Secure intake
Signed webhooks · rate limits
- 2
Identity & access
Roles · one workspace per client
- 3
Data protection
Sensitive fields masked
- 4
Knowledge
Only your approved sources
- 5
Reasoning
Claude · hands off when unsure
- 6
Actions
Approvals · no double runs
- 7
Audit
Append-only · hash chain
Works with the tools you already use
Google Calendar/Twilio/WhatsApp Business/QuickBooks/Stripe/HubSpot/JobNimbus/Boulevard/Mindbody/Weave/Toast/Mews/Slack/Wealthbox
Services
Three ways to work with us.
Agent Architecture Audit
Timeline · 3 weeks
A three-week review of your data flows, current AI use and risks, measured against the NIST AI RMF. You get a roadmap your CTO and compliance team can both sign.
Governed Agent Implementation
Timeline · 8-12 weeks
A production agent with an audit trail, role-based access, data masking, a curated knowledge base and human approval for sensitive actions. Web, WhatsApp Business, email and Slack. Built on Claude.
AI Governance Setup
Timeline · 6-8 weeks
An inventory of every AI tool you use, a NIST AI RMF profile, written policies, a vendor review process and an evidence kit for your own SOC 2 work.
How we work
Audit, build, govern.
- 01
Audit
We map your data flows, find the weak points and agree on what the agent may and may not do.
- 02
Build
We build on Claude with role-based access, data masking and human approval for risky actions.
- 03
Govern
Every month we review the logs, tune the agent and keep the controls current.
Why Loucells Core
Why teams with compliance requirements work with us.
Small team, strict process
You work directly with the people building the system, and every change goes through review and leaves a record.
Built on Claude
We use Anthropic's commercial API, and your data isn't used to train their models.
Run by us, controlled by you
Your data and your keys stay yours. The monthly retainer keeps the controls aligned with the NIST AI RMF and supports your own SOC 2 work.
Free call
Tonight at 9:47, someone will text you.
Let's make sure someone answers. The call takes 30 minutes, there's no sales pitch, and you'll leave with next steps whether or not you work with us.