Unity AI Gateway: How to Govern Multi-AI Access Across Your Company

Summary

Unity AI Gateway on the Databricks Data and AI platform has expanded beyond model routing to govern MCP servers, agents, and skills through Unity Catalog, providing a single control plane for managing model access, tool invocations, and agent behavior across an entire organization.
An agent observability dashboard shows token usage by agent and type, a performance tab surfaces success rates, latency, and error analysis, and a cost dashboard provides 30-day spend breakdowns by agent, user, and model family—all with per-team budget controls and individual overrides.
Built-in guardrails including PII detection, prompt injection blocking, and partner integrations protect against adversarial attacks, as demonstrated live when a jailbreak attempt against the governed SoulHaven Mutual insurance scenario was blocked in real time.

Unity AI Gateway: How to Govern Multi-AI Access Across Your Company

As coding agents like GitHub Copilot, Cursor, and Claude Code become embedded in daily development workflows, enterprises face a new challenge: how to govern and audit AI-assisted code generation at scale. Without centralized visibility, organizations risk exposing sensitive data, losing track of usage, and failing to meet security and compliance requirements.
In this video, we'll show how Databricks enables end-to-end governance for coding agents across the enterprise. You'll learn how to enforce fine-grained permissions, track usage and costs across teams, and capture complete audit trails of prompts, responses, and tool interactions. We'll also cover how to apply consistent policies across providers — so developers can move fast with coding agents while maintaining the control, visibility, and compliance production environments demand.
You'll walk away with the patterns to unlock coding agents safely across your organization — without creating a new compliance liability.
Talk By: Ahmed Bilal, Staff Product Manager, Databricks ; Kevin Stumpf, Senior Staff Engineer, Databricks ;

Chapters

FAQs

What is Unity AI Gateway on Databricks?

Unity AI Gateway is a centralized governance and observability layer within the Databricks Data and AI platform that controls access to AI models, MCP servers, agents, and skills across an organization. It provides fine-grained permissions, usage and cost tracking, complete audit trails of prompts and responses, and policy enforcement that applies consistently across providers.

How do I track AI cost and usage by team in Databricks?

Unity AI Gateway includes a cost dashboard that shows 30-day AI spend broken down by agent, user, and model family, making it easy to identify cost drivers and enforce budgets. Teams can set per-team spending limits with individual user overrides, ensuring AI usage stays within financial guardrails without blocking productive work.

How does Unity AI Gateway protect against prompt injection and jailbreak attacks?

Unity AI Gateway includes built-in guardrails for PII detection, prompt injection detection, and partner security integrations that evaluate each request before it reaches the model. This video demonstrates a live jailbreak attempt against a governed insurance scenario that is blocked in real time, showing that guardrails apply at the gateway layer regardless of which model is being used.

Can Unity AI Gateway govern coding agents like GitHub Copilot or Claude Code?

Yes, this video describes and demonstrates how Unity AI Gateway can govern coding agents—including GitHub Copilot, Cursor, and Claude Code—used by developers across an enterprise. The gateway enforces policies, tracks usage, and logs all interactions so organizations can give developers broad access to coding AI tools while maintaining the compliance and audit trail that production environments require.

Full transcript

[00:08] All right, everyone. Thanks for coming. I'm Kevin. I'm one of the engineers on Unity Gateway and I'm here with Bill, the PM of Unity Gateway. And we're excited to walk you through a demo of what the product looks like. Um, but before we do that, I just want to set
[00:21] the stage really quick um with an example of what a AI first insurance company today would look like that is going all in on AI. And we're going to take a look at the governance challenges
[00:33] that they're facing as they're going to roll out agents more and more aggressively. And then Bal is going to walk you for about 20 minutes through a live demo of the product that shows you how this
[00:46] company would use Unity Gateway to govern all of their MCP servers, models, and agents in the organization. So, let's meet Soul Haven Mutual, our nationwide insurance company that sells
[00:58] everything from auto insurance to life insurance and home insurance. And of course, Soul Haven is using AI already across the board for a variety
[01:10] of different use cases like coding agents. No big surprise there. From codeex, clock code to codeex and cursor. They're also using agents for automated claims processing.
[01:22] They're using agents for underwriting decisions. when our new customer wants to sign up to automatically make a decision whether the application should be approved or rejected and what the price point for it should be.
[01:35] And then finally, of course, there's also agents that are engaging with the customer to look for upsell opportunity, cross-ell opportunities, and just for customer support. But this is only what their organization
[01:49] looks like today. What we really expect to see over the coming 12 to 24 months is an explosion in agents where every employee has their own agent or many agents and then their agents have their
[02:01] own agents which will lead to the sprawl of thousands tens of thousands of agents across the organization. And of course, something like this is a nightmare for any admin and an
[02:15] organization because what you can't afford to have is to have tens of thousands of agents, hundreds of engineers and employees using those agents and AI tools,
[02:27] hundreds of different AI providers, but zero visibility. Like that's an utter compliance and regulatory and administrative nightmare.
[02:41] And what we hear all the time from our customers, and that was also a recurring theme during this conference over the last couple of days, is what enterprises really are looking for is to give their developers and employees as much choice as somehow possible. They should be able
[02:54] to use almost any model, almost any AI provider, but that needs to be done safely. And so admins need controls to govern how is AI actually used in the
[03:06] organization. And control really boils down to three different pillars for us. The first one, an extremely important one, is just observability. You need to be able to see exactly what
[03:18] every agent is doing, what every employee in the organization is doing with AI, what every token is used for because you cannot manage what you don't measure to begin with. Second, a recurring theme that we hear
[03:32] about all the time is cost controls. We want to be able to allow our organization to move as quickly as possible with AI and take advantage of it, but we can't have the increase in AI
[03:44] costs outrun the increase in our revenue or we're going to bankrupt the organization. So, we need cost controls. And then finally, we need guardrails. We need to have the ability to control what
[03:56] data an agent has access to and what tools an agent is allowed to to execute in order to stay compliant, in order to take the appropriate risks for the
[04:08] different use cases. And this is exactly what we set out to solve with data bricks Unity AI gateway. Now, some of you may know that data
[04:21] bricks has actually had an AI gateway for the last two years. It's baked into the workspace of data bricks and it's been used by thousands of customers
[04:33] really successfully. We're at this point processing on the order of 50 trillion tokens a week. And to put that quickly into context, if one can believe publicly cited numbers, that's roughly
[04:45] on the order of 20% of the traffic that Google is processing. So this isn't just a little blip in the ocean. This is already used at massive scale. It's battleh hardened infrastructure that our
[04:58] enterprise customers are using to power their AI applications reliably. And this data bricks AI gateway that we've had for the last two years has been used by our customers to govern
[05:10] models and model providers. Whether those model providers are internal to data bricks with models that we come out of the box with or external model providers where you can bring your own key to bedrock or Azure foundry and just
[05:24] govern it through data bricks. But one of the big announcements this week was that we've expanded what Unity AI gateway is able to govern. Now we've expanded it to allow you to govern the
[05:37] full gambit of AI assets to include now MCP servers, agents, and skills. And of course, given that we're now governing a ton more assets in the
[05:50] organization, it only made sense for us and our customers to integrate this with our existing governance foundation. And that is Unity Catalog, which has only
[06:02] been used to govern really data so far. Now, through this integration, we're able to govern all of these AI assets through Unity Catalog. And what that integration now gives us
[06:15] is all it solves all the observability needs. You get a catalog where everything is registered to begin with. Every agent, every MCP server, all the tools, all the models, all your external
[06:27] AI providers. You just have one control plane, one centralized control plane where you see what's going on and what's being used in the organization. You get audit logs so you see exactly what's being used by whom.
[06:39] We just recently launched a unified trace table as well that allows you to log away every single token that flows through the gateway, no matter from which agent, no matter to which model or which provider in one centralized Unity
[06:54] catalog table that you can use to mine for insights and information to how AI is used in your organization. You get a full lineage view that shows you how agents are connected to models and model providers and tools.
[07:08] You also got operational health metrics to just see how reliable the system is. Beyond that, we added Unity AI gateway budgets. Now they give you full cost controls to actually configure alerts
[07:22] where you get an email if more than X dollars a month are spent. And if you want to, you can also configure hard spend caps where traffic actually gets throttled for different users if they've exceeded their daily or monthly budget.
[07:36] And this is a fairly customizable solution where you can have different sets of groups. Like maybe you have an engineering group that gets a thousand bucks a month and then you've got power users, they get 5,000 bucks a month. And you can shift people around between
[07:48] different groups depending on the the value of their use case they're working on. And then finally, we've innovated and spent a ton of time in this area that we
[08:00] call contextual policies. And contextual policies allow you to govern of course access to individual models and tools. Um they allow you to apply guard rails at scale that take a
[08:14] look at the tokens that are actually flowing through the models and through the tools. And those content guardrails they can be things like where we come out of the box with PII reduction, PII detection, uh
[08:27] prompt injection detection, all that type of stuff. Um, we also allow you to connect to partner providers, offguard rails like the crowd strikes and zcalers of the world.
[08:40] And this entire contextual policy system is built on a fully customizable platform. So you don't have to use any of our partners or out of the box guardrails. You can provide your own custom guardrails in the form of either
[08:53] code that is fully deterministic and under your control. If you just want to have a simple reax expression or something like that, that's super easy to do. Or if you want to have and use the power
[09:05] of LLMs, you can also now use LMS as judges that take a look at all of the tokens that are flowing through your tools and through your models to decide whether you actually want to allow those tokens to go through, whether you want
[09:17] to automatically reject them, whether you maybe want to override and mask some information or whether you want to require that a human is in the loop to actually approve that action.
[09:33] And what all of this now means is that through this integration Unity catalog, you now have one centralized foundation to govern all of your data and AI. Previously, it had been entirely used for the governance of data. The AI
[09:46] assets were mostly governed actually in the different workspaces, which was a little bit of a pain if you used multiple workspaces in your organization and really needed to govern AI much more at scale. And now with this integration,
[10:00] you're able to govern really agents and models and tools at the same scale that you've been able to govern them, govern tables at scale with Unity Catalog.
[10:15] And before we take a look at the product, there's just one last thing I want to mention because it came up a little bit or pretty frequently in conversations throughout this week. I want to highlight that and emphasize that UNAI gateway is an open gateway.
[10:29] You can use it to govern AI assets that run inside of data bricks or outside of data bricks. So even if you use absolutely no model on data bricks, use no MCP server that's hosted on data bricks, no agent that's hosted on data
[10:41] bricks and everything runs outside of data bricks like you bring in your own bedrock key, your Azure foundry key, agents that run anywhere else or MCP servers that run anywhere else. It doesn't matter. You're perfectly fine to
[10:54] do that with Unity AI gateway and just use it as a registry as a system to govern access and permissions to apply guardrails or to simply extract the tokens and store the traces in a unified
[11:07] table that you can use for analysis later. And of course, it goes without saying that the gateway also works with all of these AI assets um that are running inside of data bricks.
[11:20] And with that, Bal is going to actually show you how to use the product.
[11:40] All right. Thank you, Kevin. So, you all know the problem. Soul heaven. Couple of agents today, thousands coming tomorrow. Three things that breaks. No visibility into what's running. no control over what it costs and no way to stop agents
[11:52] from taking actions they're not supposed to take. So I'll be showing you how you can solve all these three live right now in soul heaven. So I've been hearing about AI gateway all the summit. So let
[12:04] me actually show you what the product looks like. So on Jerobics you can go to AI gateway and this is where all your geni state
[12:17] lives. So here you have your models across all your providers, your database managed models as well as those coming from external providers.
[12:34] Then here you have your MCPs servers that are running on Terraix as well as external servers and those coming from SAS providers. They all live here. And soon we are adding skills. So you
[12:46] will be able to review, approve and share skills across your organization. And then we have agents uh including external agents. They can also now be
[13:00] cataloged and governed in a consistent way with AI gateway. Now one of the first thing that we hear from lot of you is how do I know how many agents I have? Who can access them? What models are
[13:14] they using? Once you have everything in AI gateway, all traffic and data automatically lands in Unity catalog tables and then you get
[13:26] this out ofthebox dashboard which gives you a live inventory usage and cost stats.
[13:40] And this is all SQL queryable tables. So you can customize this dashboard however you want and Soul Haven exactly did that. They took this
[13:52] dashboard as a starting point and then customized it to meet their requirements. So let us actually together look at that dashboard.
[14:10] All right. So this is what that dashboard looks like. The first tab is all about what's running. So on the right side you can see decisions, tokens and active users over the last 30 days.
[14:26] On the left side you can see top agents by token usage. And below you can see assets by types. So you have your agents, models, tools.
[14:38] They have close to 16 agents in production. And then below you can break down these agents by categories and domains.
[14:52] And further below here you can see what models are these agents using. Cloud, OpenAI, Gemini. But wait, I also see Quen. Now Quen is not even allowed in Salt Heaven.
[15:06] But seems some users and agents are actually able to query Quen. So we just identified some shadow AI. What can we do about it? Now with AI gateway, you can actually do very fine grain granular
[15:19] permissioning on these models. Very easy to do that. Let me go back to AI gateway here.
[15:43] permissions. Now, one thing I want you to notice is that this looks very similar to how you have been doing permissioning on your tables and data assets. It is because AI gateway is now part of Unity catalog. So you can govern models, MCBs, agents the
[15:58] same way you have been governing your data assets. So I can come here, revoke access for all users.
[16:13] But maybe I need to add my research team back because they probably want to experiment with this model. So I can only add them back here. Awesome,
[16:34] great, problem solved. But maybe not. There are like probably lot many Quen models in the catalog and there will be new Quen models coming next week. So I cannot be going into each models and doing permissioning one by one.
[16:47] So I'm kind of really excited to tell you this that you can now use AI gateway to define aback policies to do permissioning at scale. So a policy can look something like
[16:59] this. And what this policy is basically saying is that no one can access Quen models except my research team. Now this will apply to all Quen models
[17:14] as well as any new Quen models that gets added into the catalog and you can write this in SQL but you can also do the same thing uh in the UI as well.
[17:32] So you can come here and come into policies and can define your policies here as well. All right. So now you you have your policy in and this policy will grow as your catalog grows. Right? Now actually
[17:45] let's look at how these agents are performing. So we switch to the next tab performance. So you see three metrics on the tab. You have your success rate, total requests and error
[17:59] rate. Then below this error rate is broken down by agents and error type. And further below you can break down
[18:12] this error rate by models and MCPS. And you can go even more granular and looks at how the latency profile looks like for both models and MCPS. Now I as kind of I was like walking through this
[18:24] dashboard, I noticed that one of the agent is getting a lot of errors. So I can here just open Genie and ask why
[18:38] my underwriting agent is getting lot of errors and I actually just run it like just before this session
[18:52] and this is what Genie told me that this underwriting agent is actually using two models GPT and clawed and it kind of seems like all of the errors are coming from that claude model with the root
[19:05] cause being that there are a lot of server side errors. So within just a minute I was able to get to the bottom of the problem and now you can actually use AI gateway to get better reliability
[19:19] by putting in fallbacks. Again, very easy to do that. You go back to AI gateway.
[19:42] click on routing, and then you can basically add your fall back. And you can fall back to a different model or you can f fall back to a same model from different provider.
[19:59] Now when my primary fails the gateway will automatically redirect the traffic to the secondary model. My agent will keep on running. Awesome. Great. So we have solved the reliability problem. Fallbacks are in. Everybody is happy.
[20:13] platform team is especially happy, but CFOs are not happy at this point. And you probably know the reason because AI spend is a black box. Change your prompt and your AI spend doubles. Developers
[20:26] can easily rack up thousands of dollars without noticing and you have to go to 16 different admin consoles to manage budgets for 16 different agents. That is why this next tab in the
[20:40] dashboard is actually getting a lot of attention. So now all AI gateway requests gets logged with dollars attached to it. So you can get an aggregated view of how
[20:55] much you are spending over the last 30 days. And then rest of the dashboard basically breaks this spend by agent, users, family and all the way down to models.
[21:13] And then you can use this data to do forecasting. So actually let's look at the forecast here. And if you are a soul haven CFO, you probably don't want to look at this forecast. They are tracking four to 5x
[21:27] over their allocated budget. And if I look here, kind of seems most of this is coming from coding agent and from some power users story of most organization.
[21:40] So observability is awesome, but you also want to be able to put some caps so that you can keep your cost under controls and change developer behavior before your next invoice lands.
[21:54] So as you can heard from Kevin, now you can use AI gateway to put budgets across all type of agents. Let me show you how you can set one up.
[22:14] All right, I will go back to gateway here.
[22:29] Um, go to go budgets. Click on manage budgets. This might ask me to log because right now only admins can define budgets.
[22:42] All right. Create budget. I'll call this one coding agent budget. You can scope it down to all workspaces or to a particular
[22:54] workspace. And then you can put aggregated limit or a per user limit. I'll just do per user limit here. and
[23:06] we'll give 5,000 bucks to all my users. And then when that budget hit hits, I want to block usage.
[23:20] All right, let's go ahead and add this thing. But I'm thinking that maybe I need to give like a higher budget to my engineering team. I can also do that. I'll do an overwrite here.
[23:39] and I'm going to give them 10,000 bucks. So now all my users in the engineering group would get $10,000 per month. Once the budget is live, you
[23:52] can go to budget and in real time see how much runaway each user has left. Great. And if and when someone hits
[24:04] their budget, you can always come back to this page, review their usage and do an individual override and unblock them. And so now you have one view where you can come put a budget that applies
[24:17] across all type of agents including all coding agents and custom agents. Right? So we have now cost under control
[24:30] and we have fixed the reliability problem as well. But there's one thing that probably keeps lot of you up at night cuz you know cost can hurt your budget, reliability can hurt your SLA but security hurt your credibility it
[24:43] hurt your customers and if you are in an insurance business it can actually take you out of it out of that business. And that's why the this last tab is probably the most important tab for many
[24:55] of you out there. And so here you can see incidents detected over the last 7 days, percentage of traffic that's being monitored for security threats and
[25:07] agents with real-time guardrails. And the rest of the dashboard breaks down this these incidents by categories and map how these incidents correlate
[25:21] with agents out there. Now you all will be asking how we are even getting this data. So now all your agentic traces are getting logged into a unified trace
[25:34] table. And let me show you what that trace table looks like.
[25:54] All right, I'll go back to gateway go traces and this includes raw payloads coming from all type of assets, models, MCPS, tools, they get logged in one central
[26:06] governed table. You can click on a trace and get very granular information and filter by model, MCP or by identity.
[26:18] And this is really the foundation on which you can now do run your security analytics. And so once you have that data, you can run any type of classifiers on on that data. And the
[26:30] easiest way to do that is to actually use lakewatch. So lakewatch is databick security and event management software. And in just like couple of clicks, you can actually configure it to run
[26:43] advanced detections on it. So you go to Lakewatch using the app switcher.
[26:57] You can configure your data sources here and then define your rules here. And so these rules could be anything from prompt detection to PII detection to
[27:09] safety content moderation. You can bring in external tools or write your own tools. And once you have your tools, Lakewatch will run these tools on incoming stream of data.
[27:23] Incidents would be automatically picked up and put in the security cases tab. Your analyst can come to this tab, review them, triage them and resolve them. So actually let me just open one
[27:37] of the incident here.
[27:52] All right. So, this one detected a PII in my input. And one other thing I want you to notice is here is that this this incident was detected little bit after the agent has actually
[28:04] exposed the PII credentials. And so while this visibility is great, you know, it puts you in in a great footing, you get you are able to detect incidents, but the incident has already
[28:16] happened. So your injection is already in, your data is already exposed. And so what you really want to be able to do is to prevent these incidents from ever happening.
[28:28] And that's why we're kind of really excited to tell you this that now you can use contextual policies to define these real time guardrails. And if I go back to my dashboard, you would actually
[28:42] see that um lot of my incidents are coming from my coding agents. So I will show you how you can now define these contextual policies using using coding
[28:55] agents as an example. So we switch to terminal here, right? And I'm I'll be using omni agent to connect my developer machine to AI
[29:09] gateway. So I can do for example omni clot and what is bas what this is doing is basically connecting my clot code to AI gateway.
[29:27] So this is the same clot code that you use every day. But if I do model, it's basically pulling the list of models that uh my admin has configured on AI gateway.
[29:40] And for example, if I say what is this repo about, all this traffic will actually now flow through AI gateway.
[29:58] All right. So, and actually you can go back to the trace and even look at um this traffic in real time.
[30:25] I think this is the one I just sent. Oh, I got clicked on the model. Uh, but here's the here's the trace. Actually, this is a different one. Anyways, it just shows up here. It it
[30:38] just takes like a minute or two for the traces to show up, right? Actually, or maybe it's still running. So, we'll give it like couple of seconds.
[31:18] All right. that I do see tokens going up and some processing happening. Okay, we'll we'll create a new terminal and try one more time. All right, so
[31:31] omni clot.
[31:44] Uh we'll skip the basic prompt here. Let's switch model. just choose a smaller model here and we're going to say um print secrets from my
[31:56] repo config. All right, cool. So, gateway basically blocked the traffic. It didn't let the traffic even reach the model. And it
[32:08] seems like my admin has defined some sort of a policy that prevents the agent from printing secrets. So how do you even go and define these policies? So these policies are defined on the
[32:21] gateway side. Um again very easy to do that. You go to your model, go to policies here
[32:35] and then you can click on a new policy. There are bunch of out ofthe-box policies that we provide here but my favorite one is the custom one where we can you can basically express your business policy as a prompt and that's
[32:50] exactly what what's I have done here. Now I would confess that's a easier problem to solve. The harder problem is to control agent actions. And there's a bit of a tension here because the best
[33:04] agents are the one that acts on your behalf. They know your credentials. They have your access. They they have your context. So they can actually get meaningful work done. But that's also what's make them dangerous.
[33:18] You have judgment. You know where the lines are. Agents have your access, but they don't have the same judgment. And that's why really what you want to be able to do is to scope down the
[33:30] permissions of your agents while also making sure they can act on your behalf. And these sort of like complex policies is something now that you can define
[33:42] with contextual policies. So I'll go back to my terminal here. I'll clear this conversation. We'll pick a slightly
[33:56] um um slightly powerful model here. And what I want to do is to make a change into one of my repo. I'll change one of a very critical business critical
[34:08] value that uh can impact a lot of customers. And so to be able to do that, one of the first thing I need to do is to um you know configure MCPS. Um, I can
[34:21] just do omni configure MCP and this basically pulls the list of all the MCPS that have been configured by my admins. Um, so let's have the
[34:34] databick SQL MCP and then I need the GitHub MCP. Uh, and once my MCP is here, all my coding agentic harnesses would be able to use it. All right, so I can go back
[34:48] to Claude here.
[35:00] And I'm going to ask it to make a change to this renewal markup file and push that change to a main repo. This going to this is will take some time. So actually let me show you what what type of policy we have imposed. All right. So I'll go back to
[35:13] GitHub. Uh I'll go back to AI gateway MCP uh GitHub policies
[35:29] and so this is a policy that has been defined by my admin. So it's a very rich policy and what this basically policy is saying is the agent can access GitHub using my credential but the agent cannot
[35:43] delete files or it cannot make changes to a protected set of files and folders. So this is kind of very very powerful uh where you can really define these
[35:56] contextual policies that can scope down the agent permission and then you can attach this policy to an object or to a schema or to a catalog.
[36:09] All right, let's kind of see. H still running
[36:28] and just just to keep on going. This is this is this is something which you should expect. Um uh so the agent will try really hard basically by the way to circumvent the guardrails but it will not be able to do that because we have a policy that we
[36:40] have applied directly on the MCP. So it will at the end um give you something like this. All right. So we went from logging to detecting incidents to making sure like
[36:54] these incidents don't even happen. All right. Let's now quickly recap.
[37:10] Right. So, Soul Haven onboarded to AI gateway. All their assets, models, MCPS, agents are cataloged and live in one view. And just with this one step, they were able to get three things.
[37:22] Observability. They can see how many agents they have, who can access them, what resources they are using. Um they can identify shadow AI, diagnose
[37:34] reliability issues and can act on those. Uh number two, cost. They can see how much every agent, every user or every model is spending across all agents
[37:46] across and across all providers in one view. And then they can set budgets with real circuit breakers. And third is security. All data now gets locked into a central table where they can run
[38:00] classifiers with lake watch and then they can enforce real time guard rails using contextual policies at the boundary. So attack attack comes in gateway stops it before it even reaches
[38:12] the model. Yes. So that is what really is possible now with Unity AI gateway. So, Soul Heaven is a demo and is a purely fictional company, but what you just saw
[38:26] is real and is shipping over the next few days. So, you can go back and next week try what we just showed you in your workspace and on your agents. And it
[38:40] starts really by registering all your models and MCPS in AI gateway. on and on that note we'll wrap up the session.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.