Piquant Book a pilot

How it works

A gate on every machine, policy from one place.

Piquant has three parts. An agent on each machine checks every call. An optional gateway serves shared company tools. A control plane on your own server holds policy, relays approvals and keeps the record.

The parts

Where Piquant runs

A PERSON'S PC AI tool Piquant agentchecks every call Piquant appstatus, approvals GATEWAY (OPTIONAL) Piquant gateway: shared tools Control planeyour server: policy, approvals,records, console, kill switch COMPANY TOOLS (MCP)GitHub, Jira, databases,internal APIs ALSO THROUGH THE GATEWAYclaude.ai, ChatGPT,automated agents, web apps Microsoft Entra IDwho people are, their roles
Green arrows: governed calls to company tools. Grey arrows: policy, approvals and records between the parts.

Agent, on each machine

Runs once per Windows or Linux machine as a system service. It knows who is calling from their sign-in to the operating system, applies their policy and adds their keys.

Piquant app

Shows each person their status, what they can use, requests waiting for them and the AI tools on their machine. It asks approvers to decide.

Gateway, optional

Serves shared company tools from one server, with company keys that never reach a laptop. claude.ai, ChatGPT, Copilot Studio and automated agents connect here.

Control plane

On your own Linux or Windows server: signed policy, approvals across machines, the record of every decision, an admin console and the kill switch.

AI-tool modes

Three modes, set per person by policy

The mode decides how far Piquant reaches into a person's AI tools. It follows their Entra role.

ModeForWhat happens
FreePeople evaluating AI toolsPiquant shows what their AI tools use but changes nothing. Calls through Piquant still follow policy.
ProvideMost peoplePiquant adds the approved company connections to their AI tools. Their own connections keep working, with a warning.
LockPeople who work with sensitive systemsTheir AI tools may use only connections that go through Piquant. Every call follows policy and is recorded.

Policy

Rules written once, signed, and followed everywhere

A rule names who it covers (an Entra role, a group or a person), which tools on which servers, and the effect: allow, deny, or allow after an approval.

Policy is prepared in the console, reviewed, and signed on an administrator's workstation, never on a server. Every agent checks the signature before it follows a rule, so neither a compromised server nor anyone between them can change what people may do.

Anything no rule allows is denied.

A rule, in words

Standard users may read issues and code on GitHub. Merging needs a team lead's approval. Deleting a repository is denied for everyone.

When something goes wrong

Stop it within seconds, and know what happened

Kill switch

Suspend a person, a tool or a device from the console, the command line or a security playbook. Calls stop within seconds; the person's Piquant app says why and until when.

A record of every decision

Who called which tool, the decision and the rule behind it, in a chain where a removed or changed entry shows. Never the content of the call.

To your SIEM

Records go to your own control plane, and from there through OpenTelemetry to the tools your security team already uses.

Tamper checks

Agents report the state of their installation. A configuration changed by hand on a machine is flagged in the console.

Start with one team.

A pilot begins in provide mode: people keep their AI tools, and Piquant adds the governed connections. Locking, approvals and shared company tools follow when the team is ready.