Beta We're shaping PromptLayer with early users. Join the beta feedback Discord →

New Runtime Intelligence

Understand your
AI systems.

Trace requests, master your token economics, attribute AI spend to features and customers, and identify optimisation opportunities across every AI workflow.

Start free → View docs
One-line install PHP · TypeScript · REST No credit card
orbit-labs/support-bot/Intelligence production Last 30 days
Intelligence Workflows Costs Attribution Requests
Estimated spend · 30d
$1,284.50 18.2%
vs $1,086.30 previous 30 days
Top driver
draft_reply
55% of spend · +31% MoM
Recommendations
3$246/mo
potential monthly savings
Data confidence
High
98% pricing · 86% attribution
Why spend changed last 30 days 87%
Spend rose +18% — almost entirely from the draft_reply workflow.
Impact +$198/mo Driver draft_reply Model gpt-4.1
105  executions this period
+31%  volume vs prev 30d
$0.0021  avg cost / request
55%  of total project spend
View workflow Compare features
Recommended action switch model 82%
Move draft_reply from gpt-4.1 to gpt-5-nano.
Outcome −68% cost Save $134/mo Risk Low
Similar token profile
Similar request complexity
Historical latency acceptable
92%  output overlap on replay
Test comparison View analysis
Spend by feature View attribution →
draft_reply$706 · 55%
summarise_thread$283 · 22%
classify_intent$180 · 14%
other$116 · 9%
Data quality Details →
Pricing coverage
Models with known unit pricing
98%
Attribution coverage
Requests tagged to a feature
86%
Estimate confidence
Weighted across all spend
91%
Works with OpenAI Anthropic Custom integrations
The problem

AI systems become difficult to understand surprisingly quickly.

The first AI request is easy. The hundredth isn't.

Once workflows involve multiple models, prompts, retries, tools, customers, and teams, simple questions get hard to answer:

01 Why did AI spend increase? →
02 Which workflow costs the most? →
03 Which customers generate the most cost? →
04 Which team owns this spend? →
05 What should we optimise first? →

P PromptLayer is built to answer those questions.

The platform

Observability and cost intelligence,
in one system.

PromptLayer sits between AI observability and AI FinOps — connecting what your system does with what it costs and who owns it.

01 Runtime Intelligence

Know what changed and where to look.

Surface spend increases, growth drivers, and recommendations in one place — each with a confidence indicator so you know how much to trust it.

What changedWhy it changedConfidence
02 Workflow Visibility

See complete execution paths.

Understand how requests move through multi-step AI systems. Visual waterfall views reveal exactly where time and cost are spent.

Nested spansTool callsRetries
03 Cost Intelligence

Track estimated spend, everywhere.

Follow estimated AI spend across models and providers, then break it down by workflow, feature, customer, team, and cost centre.

Per-modelPer-featurePer-provider
04 Attribution

Understand who owns spend.

Allocate costs to products, customers, teams, and features. Move beyond raw token counts and into real accountability.

By customerBy teamBy feature
05 Recommendations

Find what to optimise first.

Identify opportunities to reduce spend with quantified savings estimates, and investigate the workflows driving the most cost.

QuantifiedPrioritisedEvidence-backed
06 Data Quality

Know how much to trust the numbers.

Track pricing coverage, attribution coverage, and confidence levels so you always know how trustworthy your spend data is.

PricingAttributionConfidence
From traces to answers

PromptLayer explains spend — it doesn't just report it.

The same data behind your traces becomes answers to the questions teams actually ask.

1
Why spend changed
Growth drivers surfaced automatically, with the evidence behind them.
2
Which workflows drive cost
Ranked by share of spend, not just request volume.
support-bot / Costs / By feature
Spend by workflow last 30 days
Workflow Owner Requests Share Est. cost
draft_reply Support · Maya 9,420 55% $706.10
summarise_thread Support · Maya 6,140 22% $283.40
classify_intent Platform · Dev 11,280 14% $180.20
enrich_profile Growth · Lee 2,310 9% $114.80
Recommended: move draft_reply to gpt-5-nano — est. −$134/mo, 82% confidence.
What to do →
3
Which customers generate spend
Costs allocated to owners, teams, and customers.
4
What should happen next
A concrete recommendation with quantified savings and confidence.
How it connects

Built for modern AI systems.

PromptLayer connects operational AI activity with cost and ownership information — so teams can understand both technical behaviour and business impact in one place.

Captured Request Every model call, wrapped at the SDK. Operational →
Grouped Workflow Spans assembled into execution paths. Operational →
Resolved Model Provider, model, tokens, latency. Operational →
Priced Cost Estimated spend per request. Business →
Allocated Attribution Tied to feature, team, customer. Business →
Suggested Recommendation A prioritised action to take next. Business
Operational activity Business impact
What teams do with it

Answer real questions about your AI.

Not dashboards to interpret — outcomes teams reach with PromptLayer today.

Identify expensive workflows
Rank workflows by share of spend and find the few that drive most of the cost.
Track spend across teams
Break estimated cost down by team and cost centre, not just by model.
Attribute costs to customers
Allocate AI spend to the customers and products generating it.
Understand AI usage trends
Watch how request volume, tokens, and cost move over time.
Investigate spend increases
Trace a cost spike back to the workflow, model, or customer behind it.
Improve confidence in reporting
Know how much of your spend data is priced, attributed, and trustworthy.

Understand what your AI systems
are actually doing.

PromptLayer combines workflow visibility, cost intelligence, attribution, and recommendations into a single platform for AI teams.

Start free → View documentation