IndyKite (@Indykite) header
IndyKite (@Indykite) profile picture

Enterprise AI needs trusted data. IndyKite brings trust, visibility, and control—connecting and governing data to power intelligent applications

Joined July 2021
How do you keep an AI agent within the scope of the task and still give it the flexibility to get it done? 🤖 Intent-based access control does it by carrying the purpose of the original request through the workflow. That limits agent actions to what the task requires. Join us for Beyond Delegation: Intent-Based Access Control for AI Agents on Monday, November 30 at 4:00 PM CET (10:00 AM EST | 7:00 AM PST), where we bring it to life with a live demo. Save your spot 👇 hubs.ly/Q04FVG8n0
13
An agent asked to check an account balance shouldn’t be able to trigger a payment three hops later. Yet when agents inherit a user’s full permissions, that’s exactly what can happen. Our new E-Guide introduces Intent-Based Access Control (IBAC) - a runtime framework that locks an agent strictly to the task it was assigned, so it can’t misuse tools it was never meant to touch. Get the E-Guide 👇 hubs.ly/Q04FMTrV0
16
Ask an AI agent for one thing and it can inherit all of your permissions. Here’s how to limit it to what you asked for with Intent-Based Access Control (IBAC) in IndyKite AgentControl 🚀 Learn more about IBAC here 👇 hubs.ly/Q04FDHsv0
60
BIG NEWS: Intent-Based Access Control (IBAC) is now available in IndyKite AgentControl 🚀 IBAC turns what a person asked for into a structured intent, carries it across every agent and tool in a multi-agent workflow, and enforces it at every step. Every agent gets exactly the access its task needs, and nothing more. Key highlights: 🔸 Least privilege for every request: AI translates the request once, using OpenID AuthZEN semantics. After that, rules make every decision. 🔸 Carried across every agent: a signed IndyKite token travels agent to agent and agent to tool, across A2A and MCP, and downstream intent can only match or narrow the original. 🔸 Grounded in your data: intent is expressed in your own knowledge graph, and access to data is decided on relationships, consent, sensitivity, provenance and trust. Check out the full announcement here 👇 indykite.ai/news/indykite-la…
1
1
40
IndyKite retweeted
You gave your agents access to the knowledge graph...what if someone asks one to pull the HR's sensitive data? 🔐 Security is sometimes skipped when building knowledge-based AI. And when they do think about it, role-based access control tends to lead to role explosion and over-permissioning, which is a bad fit for agents whose behavior is dynamic and hard to predict. We wanted to share this livestream from a few weeks ago that we are sure you'll find interesting: Alex Babeanu covers why connected data raises the security stakes, why RBAC breaks down, and how your own graph can become a policy engine your agents can trust. Full session:bit.ly/4BUvuOe #Neo4j #AgenticAI #AISecurity #KnowledgeGraph
4
5
23
1,877
We just open-sourced the Intent Agent to set a standard for how AI agents express intent 🚀 Watch how it works: a plain-language request in, a structured, standards-based intent out. Available now on GitHub 👇 hubs.ly/Q04FbN-Q0
1
2
113
AI agents need access across tools, data and systems to deliver value. That access also means an agent can drift beyond the task, accessing unrelated data or making changes you never asked for, all within your permissions 🤖 Intent-based access control keeps agents within the scope of the task while giving them the flexibility to get it done. On Monday, November 30 at 4:00 PM CET (10:00 AM EST | 7:00 AM PST), we bring it to life with a live demo 🚀 Register now 👇 hubs.ly/Q04F4y8c0
1
24
AI agents are getting the keys to everything, even when the task only needs a fraction of it. Here’s how to scope each agent to its purpose with the open-source IndyKite Intent Agent 🚀 Available now on GitHub 👇 hubs.ly/Q04DZtpb0
2
1
142
Today, we’re excited to announce the open-source release of the IndyKite Intent Agent - setting a new standard for how AI agents safely express and enable enforcement of intent! 🚀 The Intent Agent captures the intent behind a user’s request and expresses it in a structured form that authorization rules can evaluate. Built on OpenID AuthZEN semantics, it provides a basis for checking agent actions against the original task. Key highlights: 🔸 Standards-based: Built on OpenID AuthZEN & Agent2Agent (A2A) protocols. 🔸 Deterministic security: Rules make the decisions, not probabilistic AI hallucinations. 🔸 Identity-verified: Bound to verified user IDs, immune to prompt-injection impersonation. The future of multi-agent workflows is intent-driven, auditable, and open. Check out the full announcement here 👇 hubs.ly/Q04DRVDM0
1
2
78
What if an AI agent only saw the tools that match what the user asked for? Say someone asks an agent to book a flight. A gateway in between checks each tool against that request. A tool that looks up customer details doesn’t fit, so the agent never sees it, even if it’s authorized to use it. Learn how that check works 👇 hubs.ly/Q04DKVSp0
1
20
How IndyKite works, in 2 minutes 33 seconds 🚀 Follow an AI agent’s request as IndyKite checks it against live context (data source, sensitivity, the agent’s task) and decides in real time what the agent can know and do. Check it out 👇 hubs.ly/Q04DyYhB0
22
You’ve let AI agents into the enterprise. Are you ready to manage them? Today in San Francisco, our Founder & CEO Lasse Andresen @forgerock1 is on stage at (RBLN) West with Jorge Zelaya, CISO at Atmosera, for “Uninvited Guests & Unchecked Agents”, on keeping AI agents useful without putting the enterprise at risk. If you’re there, don’t miss it. More details 👇 hubs.ly/Q04DdJqB0
22
What happens when an AI agent is authorized, but the action it takes isn’t? Our own Dave Bennett takes the question to AI Builders Monthly London on Tuesday, November 10. More details 👇 hubs.ly/Q04D4jCF0
16
An AI agent can end up doing something other than what it was asked. Catching that gap means checking its actions against an intent. But whose: the agent’s own, or whoever gave the instruction? Alexandre Babeanu explains which one to go by, and why, in Intent based access control: Whose Intent is it anyway? Read more 👇 hubs.ly/Q04C-8YC0
13
Three talks, one evening, London, Tuesday, November 10 🇬🇧 We’re bringing the team down to Civo Tech Junction (6–9 PM) for AI Builders Monthly, and you definitely want to be in the room for these: Our own Dave Bennett will take the floor on what happens when an AI agent is authorized, but the action isn’t. Raffi Krikorian from Mozilla will talk about building your own agent, and Neo4j's Martin O'Hanlon will cover using AI to make sense of a WhatsApp group. More details and registration 👇 hubs.ly/Q04CS2kV0
24
Permissions that change when the situation does. Watch the full webinar with Alexandra Teachout and Sean Evers 👇 hubs.ly/Q04CS0NV0
1
15
When an AI agent acts, what should its actions be checked against? An agent can end up doing something other than what it was asked. Whether that gets caught depends on whose intent its actions are checked against: the agent’s own, or the person who gave the instruction. Our own Alexandre Babeanu makes the case for the person’s intent in Intent based access control: Whose Intent is it anyway? Read more 👇 hubs.ly/Q04CyNy-0
11