Which prompt version runs in production today, and which team pays for its calls? Relevant sets access, spend, and prompt-change policies with your leads and enforces them in your routing layer, so every call can run inside the rules you signed.
Our consultants turn your AI usage policy into rules a system can check, one for each team, application, and data class. Release sign-off for whole AI systems belongs to AI governance implementation, and the framework takes its risk tiers as input.
Relevant engineers issue a key to each application and team, tie it to your single sign-on groups, and check its rights before a request reaches a model. Models cleared for restricted data run on an on-premise AI deployment inside your perimeter.
Spend follows the budgets you approve: our engineers write a monthly limit for each team and application into the routing layer and report cost per request back to the team that sent it.
A versioned registry holds every prompt your applications run in production, and our engineers connect it to your release process with the approver your rules name.
Each approved rule becomes a check in the routing layer between your applications and the model providers, so every request meets the same conditions. Where that layer is still missing, Relevant builds it through our AI orchestration development services.
One record per call carries the caller, model, prompt version, cost, and rule decision, and our engineers map each field to the obligations that AI compliance services assign to each system.
Relevant engineers read your provider accounts, gateway logs, and repositories and list each application, key, model, and monthly spend. You approve the inventory before the rule work starts.
Workshops with your security, finance, and platform leads settle the access, spend, change, and evidence rules for each team. Your leads sign the set, and Relevant prices the build against it.
Relevant engineers break your recent provider usage down by team and application, using the keys from the inventory to assign each call. You approve each team’s starting budget against its own figures.
The first team moves behind the checks with its keys, budgets, and alerts live. With AI guardrails implementation your team can also get content checks on each prompt and response in the same rollout.
Every application now sends its calls through the enforced checks. Your platform team owns the rule code, the prompt registry, and the audit records, and Relevant stays on for rule changes for as long as you choose.
Your rules live in your own routing layer, so a change of model provider leaves every key, budget, and record in place.
ISO 27001:2022 certified. Keys and call records we handle during the build stay under controls audited against that standard.
Up to 50% faster delivery and up to 20% fewer bugs. The Relevant AI Delivery Framework™ and the AI-assisted development practice behind it speed up rule code and test cases, with every release still passing senior code review.
Relevant delivers software for fintech, pharma, healthcare, and energy clients, where call records answer to regulators as well as to internal audit.
Relevant builds production LLM applications for clients, so each rule comes from engineers who have shipped model calls to production.
96% employee retention. The engineer who wrote your budget rules stays available when a team asks for a higher limit.
Five figures set the size of the work: the applications and scripts calling a model today, the provider accounts and API keys behind them, the teams that need a separate budget, the prompts running in production, and the regulations the call records must answer to. Shared keys add work, because each one has to be split before a team can get its own budget.
If a figure is missing, the inventory stage produces it, with nothing prepared on your side. The signed rule set and a priced rollout plan arrive within 14 days of kickoff. Contact us with the figures you have, and the first call turns them into a scoped range.
Each layer answers a different question about the same request. Budgets and model rights belong in the routing layer and content filters in the guardrail layer, so each layer changes on its own schedule without touching the others. The questions divide this way:
Yes, they reach every model call an agent makes. Each agent sends those calls through the same routing layer as any application, and it holds its own key, model list, and budget.
Agents need one change to the spend rules. A single user task can trigger dozens of model calls in a loop, so a limit per request misses the cost of the task. Agent budgets sit per agent and per workflow instead.
Tool calls sit outside the model layer. What an agent may do in your CRM or billing system depends on tool permissions. Tracing each step an agent takes across those tools is the job of AI agent monitoring, which can run in the same rollout.
Vendor AI features, such as the assistants inside your CRM or office suite, send their model calls from the vendor’s systems, so those calls run outside your routing layer. Three other controls reach them: the vendor’s admin settings for which features each group may use, the contract terms on data retention and model training, and the identity rules that decide who holds a licence. Our consultants add each vendor feature to the rule register with its owner, so one list covers the calls you route and the ones you contract for.
You choose the behaviour per application in the rules workshop. A switch to a smaller model needs the closest watch, because answer quality changes while the product side sees nothing, so the call record marks each switched call for the product owner to review. The three options carry different trade-offs:
Relevant puts those decisions into your routing layer, so every call runs inside the rules you signed.
AI
AI
AI
