Premium models on routine work
Your most expensive model is writing boilerplate because nobody set a default.
Control AI cost and quality across every team, without slowing anyone down.
AI use is spreading across your teams faster than anyone can govern it by hand. Set which models, tools and budgets each team gets, once. XTrace holds every call to that policy, ties spend to the workflows it serves and recommends where to cut waste, so you can approve more AI with confidence.
· Governance overview
The problem
Most AI waste isn’t a decision anyone made. It’s a default nobody set.
Tool sprawl
Every team wires up its own tools and skills. No one knows which ones help and which ones hurt.
Quiet failure loops
Agents retry and work around broken tools. The task finishes, nothing looks wrong, and the bill keeps climbing.
Spend without an owner
You can see spend by vendor. You can’t see which workflow it served or whether that workflow got better.
Our research on how large engineering organizations govern AI spend and where it leaks.
Read the research
How it works
One control point in the request path. One console to run it.
The AI gateway sits between your agents and apps and the models they call, as a proxy, an SDK or both, with flexible deployment depending on your needs: hosted by us, self-hosted or in your VPC. It runs the policy you set on every call and analyzes what happens after.
What it learns about routing, caching and budgets arrives as recommendations a person approves in XTrace. Fixes for repeat mistakes and policy violations reach the XTrace plugin inside each AI tool automatically, and XTrace turns the results into spend and ROI reports.
Request path. Every model call from AI tools, agents and apps passes through the gateway.
Recommends. The gateway analyzes traffic and surfaces routing, caching and budget changes. A person approves them in XTrace.
Enforces. Rules and policies go to the XTrace plugin, which applies them automatically inside each tool.
Where each surface sits. The AI gateway runs the routing, caps, caching and policies you set on every call, and recommends changes for a person to approve; the XTrace plugin enforces rules and policies automatically inside each tool.
Route
Decide once. It applies on every call.
Your platform team sets the policy in XTrace. The gateway holds every call to it, and tells you where it could work better.
AI gateway
Policy held in the request path.
Sends each task type to the model your policy sets: efficient models for routine work, premium models where they earn it.
Runs the caps, caching and access policies you set, per team, workflow or task type.
Gives each team only the tools and skills it is approved for.
Attributes every call to a team, workflow and model, so every cost has an owner.
Recommends routing, caching and budget changes from what it sees. It changes nothing until a person approves.
No interface of its own. You see its effects in XTrace, in your policies, recommendations and spend reports.
Policies & routing
Where your platform team sets the rules.
Routing configuration: default and escalation model per task type.
Which tools and skills each team gets: approve, retire, roll out.
Budgets per team and workflow, with every change logged with an owner.
Gateway recommendations, each approved or dismissed by a person.
Adopt new models centrally, without every team reconfiguring.
Sample XTrace screen for governance. A routing policy table sets the default model tier, escalation tier and caching for each task type. Gateway analysis recommends moving test generation from the Advanced to the Standard tier, with 99.1% quality parity and about $7.4k a month in projected savings; it waits for a person to approve it. A matrix shows which tools and skills each team may use, with one pending access request from Data science for read access to production logs. Policies apply on every call; gateway changes made without approval: zero.
· Policies & routing
Routing & access
Policies you set here apply on every call. Changes suggested by the gateway wait for approval.
Routing policy5 of 14 task types
| Task type | Default modelDefault | Cost / taskCost | ||
|---|---|---|---|---|
| Code generation | Advanced | $0.42 | ||
| Test generation | AdvancedStandard | $0.31$0.12 | ||
|
Recommended: move test generation to the Standard tier. Quality parity 99.1% on 2,140 sampled tasks. $7.4k/mo projected
ApprovedApplies at next sync
Dismissed
| ||||
| Ticket summaries | Light | $0.02 | ||
| Incident triage | Standard | $0.18 | ||
| Data analysis | Standard | $0.27 | ||
Tool and skill accessBy team
Team
Coding agent
Web search
Prod logs
Deploy skill
CRM
Payments
Platform
Support
Data science
Marketing
-
Access request· 2 days ago
Data science: read-only access to production logs
For incident analysis. Write access stays blocked by policy.
Applies once approvedApprovedPolicy updated Dismissed
Monthly capsEnforced in the gateway
Support$11.3k of $14k
Data science$11.5k of $18k
Platform$10.3k of $22k
Team-level policies · 5 teams shownGateway changes made without approval 0
Your teams barely notice.
Engineers meet governance through the XTrace plugin, inside the AI tools they already use. It enforces team rules and company policies in the session, and offers a model suggestion or a budget note in the moment. When everything is fine, it stays out of the way.
Teams refine their own skills inside the policy you set, and nobody waits on a daily usage cap.
Coding agent · billing-service
Squash the cleanup commits and get them onto main.
Squashed into one commit. Pushing to main.
Run
git push --force origin mainHeld
XTrace plugin
Blocked · team rule
Force-pushing to main isn’t allowed in this repository.
Changes reach main through a reviewed pull request. Push the branch and open one instead.
Suggested fix
git push origin cleanupRule set by your platform teamUnderstood. Pushing the branch and opening a pull request.
Run
git push origin cleanup && gh pr create --fillDonepull request opened for review
XTrace plugin active1 team rule applied · session continued
Measure
See what your AI spend bought, and what to fix first.
XTrace maps spend to the workflows your teams run and puts it next to the KPIs you already track.
Automatic anomaly alerts flag a workflow or tool that suddenly costs more. Gateway recommendations wait on the same screen, with estimated savings where XTrace can calculate them, and a person approves each one: change routing, fix a skill or retire a tool.
Sample XTrace screen. Four recommendations from gateway analysis wait for a person to approve them. Three carry a projected saving, about $15.4k a month together: route test generation to a cheaper model tier, cache repeated repository context in code review, use the light tier for ticket summaries, and set a monthly cap on nightly evaluation runs. Nothing changes until someone approves. Side panels show a spend anomaly (Support triage at 3.4 times typical hourly spend, owner notified) and an allocation rule that assigns the payments repositories to the Payments team, with 97.2% of AI spend allocated. Team-level data.
· Recommendations
Recommendations
Suggested by gateway analysis of the last 30 days.
Awaiting approval4
Projected savings$15.4k/mo
No change until someone approves.
-
Routing· 6 teams
Route test generation to the Standard tier
Quality parity 99.1% on 2,140 sampled tasks
$7.4k/mo projectedApprovedQueued for rollout Dismissed -
Caching· 9 teams
Cache repeated repository context in code review
38% of prompts repeat the same context within a day
$4.9k/mo projectedApprovedQueued for rollout Dismissed -
Model choice· 4 teams
Use the Light tier for ticket summaries
Quality parity 98.4% on 1,310 sampled tasks
$3.1k/mo projectedApprovedQueued for rollout Dismissed -
Budget· 3 teams
Set a $6k monthly cap on nightly evaluation runs
Averaged $8.2k a month last quarter, with no cap set
No estimateApprovedQueued for rollout Dismissed
Spend anomalySince Tue 14:00
Support triage · Customer Support
3.4×typical hourly spend
Owner notified · $1,870 todayInvestigate
Allocation rule 12Active
WhenRepomatchespayments/*
OrAgent tagischeckout
Assign toPayments teamCC-4410
AI spend allocated97.2%
$3.1k unallocated this month
Team-level data · 23 teamsApproved this quarter $21.3k/mo
Team AI health
Failing tools, loops and missing context, by team.
Every finding is tracked until the fix is verified.
Skill performance
Which shared skills earn their keep.
With suggested improvements for the ones that don’t.
Rollout readiness
Before agents reach a new team or codebase.
What’s in place, what’s missing and the top three fixes.
Report
The readout, without building it by hand.
AI spend, year to date
$1.24M
+27% Q3 vs. Q2
Budget vs. actual, FY26
99%of plan to date
Full-year forecast
$1.84M
Base forecast, 2% over budget
Spend tied to KPIs
81%
14 workflows mapped
Monthly spend and forecast
- Base$1.84M+2% vs. budget
- With optimizations$1.73M−4% vs. budget
With optimizations: spend if XTrace's suggested changes, such as a model or skill change, are applied.
Cost per outcome
Six-month changeWorkflowUnit costTrendChange
Customer supportper resolved ticket
$0.84
−22%
Engineeringper merged pull request
$2.10
−14%
Claimsper processed claim
$1.12
−9%
Finance operationsper reconciled invoice
$0.31
−18%
Salesper account brief
$1.40
+6%
23 teams · 14 workflows · team-level data
FY26
Teams with steady AI use
17of 23
+9 teams since rollout
Stalled sessions
11%
From 24% before standards
Standards adopted
12of 15
3 rolling out
Work done with AI, before and after
Platform+33 pts
Customer support+25 pts
Data science+32 pts
Payments+36 pts
Mobile+29 pts
Finance operations+25 pts
Risk & compliance+17 pts
Practice adoption
Teams using each standard- Plan before generating code21 / 23
- Shared project context18 / 23
- Tests on agent output15 / 23
- Right-sized model choice12 / 23
- Reuse of team knowledge10 / 23
23 teams · team-level data
Before: Q1 FY26 · After: Q3 FY26
The same data becomes the reports a CTO’s office reads: cost per outcome, budget against actual and what changed since last period.
Clear at a glance and always current, in the app or on a schedule.
Reporting to finance too?
See FinOps for AI
Improve
Fix a failure once. Every team benefits.
A point-in-time audit surfaces the failures in your environment today, and addressing them is a sound first step. But new failure modes emerge as models, tools and teams change, so durable control depends on continuous detection and remediation. XTrace provides exactly that: it identifies retry loops and recurring errors in live sessions, resolves them automatically, and the XTrace plugin enforces each fix in every team’s sessions, not only where it was first found.
Sessions by team · 14 days
- XTrace spots a mistakeOne team's agent keeps retrying the same failing step, and other teams start to hit it too.
- It fixes it automaticallyThe fix becomes a team rule on its own. Teams that want a look first can review it.
- Every team's plugin enforces itThe XTrace plugin stops the mistake inside each session. Then XTrace looks for the next one.
One team’s repeat failure, fixed automatically as a team rule, then stopped by the XTrace plugin in every team’s sessions. Team-level only.
What runs where
XTrace fixes repeat mistakes and policy violations automatically, and the XTrace plugin enforces them inside AI tools; teams that want to can review fixes first. The AI gateway only recommends routing, model, caching and budget changes; none takes effect until a person approves it in XTrace.
Getting started
We connect your stack.
Your team takes it from there.
Connect
We connect your AI providers, agents and apps to the gateway, so every call runs through one control point.
Configure
Set routing, tool and skill access and budgets in XTrace, around the workflows and KPIs you own.
Report
Spend and ROI reports and gateway recommendations follow, in the app or on a schedule.
Add-on
Move faster with expert help.
Add a spend and stack assessment, hands-on configuration support and workshops with your platform and engineering teams.
Trusted foundations
The backing and certifications your security and procurement teams ask about.
Questions
Common questions
No. The AI gateway runs the routing, caps, caching and policies your team has set, and analyzes traffic to recommend changes. A person approves each one in XTrace. Fixes for repeat mistakes are different: XTrace makes them automatically and the XTrace plugin enforces them inside AI tools, with review available if you want it.
Approve more AI, with the control to back it.
Show us your stack and the workflows you’re accountable for. We’ll show you where the spend leaks and how to close it.