Hexamind: the factory infographic
An open-source gatekeeper for every tool an AI agent calls: the closest neighbour to what we do.
What the company does
In normal terms
French AI company ("L'IA utile et maîtrisée"); publishes Hexgate, an open-source (MIT) authorization layer for AI agents that gates every tool call through allow/deny/approval decisions by role, with signed policy bundles and an audit trail.
Hexgate sits between an agent and its tools and enforces the company's policy on each call, by role, with signed policy bundles and an audit trail.
In lay terms
Hexamind makes a bouncer for AI assistants. Every time an assistant tries to do something, the bouncer checks the list and says yes, no, or 'ask a person first'.
The risks it faces, especially from AI agents
In normal terms
- Policy says yes, business says no. A gate enforces whatever the policy says; if the policy is wider than the agent's real task, the gate faithfully allows the wrong thing.
- Approval fatigue. Human approval steps on frequent tool calls get rubber-stamped, turning a control into a formality.
- Proving it to the buyer. Enterprises and insurers need evidence that the gate covers every path the agent has, not only the tools it was wired to.
In lay terms
- Policy says yes, business says no. A bouncer with the wrong guest list lets the wrong people in, politely.
- Approval fatigue. If the bouncer asks the manager every five minutes, the manager starts saying yes without looking.
- Proving it to the buyer. The bouncer only guards the front door. Who checks the back door?
How RiskMandate helps
- Reachwhat it can actually access or affectEvery tool and data path each agent can reach, not only those behind the gate.
- Mandatewhat the business authorisedThe task each agent was approved for, written down by the business.
- Gapthe difference: the riskTools the gate allows that the mandate never needed.
- Barrierswhat really stands in the wayHexgate becomes a named technical barrier, with its audit trail as evidence.
- Provenancewho owns it, who reviewed it, which versionWho signed each policy bundle, who reviewed it, which version runs.
In normal terms
RiskMandate is the record above the gate: it says what the gate's policy should be, from the mandate, and shows the gate as evidence that the barrier holds. Hexgate enforces; RiskMandate explains and proves. A joint demo writes a policy in RiskMandate and enforces it in Hexgate.
In lay terms
Hexamind builds the bouncer; RiskMandate writes the guest list and shows the insurer the bouncer is doing the job.
A first line for the conversation: Your gate enforces the policy; we write down what the policy should be and prove it holds. Shall we try one agent together?
Sources and method
The facts in grey (what it does) are from the company map, which cites:
- www.futurefeed.to, read 2026-11-01: third-party list naming Hexamind/Hexgate among exhibiting AI startups
- github.com, read 2026-11-01: own GitHub organisation: description, website, Hexgate repository
- pypi.org, read 2026-11-01: Hexgate package: runtime authorization for AI agents
The risks and how RiskMandate helps are our analysis, written by dev@web-summit.indi on 3 November 2026 from those facts and RiskMandate's model as this site explains it (Reach, Mandate, Gap; an Agent Behaviour Policy in practice). They are not the company's own statements and name no incident. A person reviews each before it is used.