Real-time FinOps for AI development

Know what your AI built, and what it cost

AIMS ties every AI request to the project and feature being built, and enforces budgets before a single token is spent.

Request early access
AIMS · Acme Consulting
Spend overviewWhat your AI work cost this month
Attributed spend£128,490↑ 12.4% vs last month
Billable to clients£112,30087% of total spend
Requests blocked3Stopped at budget cap
Spend over time
01 Jun15 Jun30 Jun
Active stories
checkout-redesign
Northwind · £4,209.58
✓ Match
refund-flow
Northwind · £2,680
✓ Match
kyc-checks
Meridian · £5,940
Checking…! Scope
Attribution status: 96% of spend attributed
Choose active workNorthwind · Payments
checkout-redesign ReadyActive ✓
refund-flow Ready
kyc-checks Ready
Govern before spendcheckout-redesign
Story budget£5,000
Model accessGPT-5
Block requests at 100% On
Developer workflowStart governed session
$ aims run --story checkout-redesign
✓ Scoped key issued · expires in 4h
Models: gpt-5, claude-sonnet
Budget remaining: £790
Automatic attributionEvery request reaches the right story
VS Code→Gateway→GPT-5
checkout-redesign + £0.42
Proof, not tags · on the roadmapCompare claimed work with the diff
checkout-redesign
payments/* · ui/checkout/*✓ Match
kyc-checks
auth/session/* · out of scope! Review
1 · Select the work being built. AIMS connects every request to an active story.
Cross-vendorWorks in your existing toolsShort-lived keysClient-owned accountsFeature-level attribution
01 · AIMS in action

See AIMS working on a real engineering workflow

From an AI request to a feature-level cost record, governed, attributed, and traceable.

02 · The problem

Your AI coding bill has stopped being a rounding error

For a growing number of teams, AI coding spend is becoming a material line item. It climbs as agents run longer, in parallel, over bigger context, yet the person signing for it still cannot say what any of it bought.

Subscriptions hide the signal. Per-developer dashboards answer “who spent it” and stay silent on “what for”.

Four engineers, Oliver, Charlotte, Harry, and Emily, feed an illustrative £184,290 of AI coding spend. Where that spend went, across projects, features, and products, is unknown.

Who spent it

OliverFrontend
CharlotteBackend
HarryDevOps
EmilyML / AI
AI Coding Spend
£184,290
  • Agents running longer
  • Bigger context
  • Parallel runs

What it built

ProjectsSpend not traced to a project
FeaturesSpend not traced to a feature
ProductsSpend not traced to a product

That is not a reporting gap. It is a governance gap.

04 · Proof, not tags

Soon, the work checks the timesheet

A budget you can game is not a control. AIMS is built to reconcile claimed allocation against observed work (the diff, the files touched, the branch) before it reaches an invoice.

  • Live today Attribution, budgets, and enforcementLive in AIMS today.
  • Next Scope reconciliationIn design with our first delivery partners.
Claimed allocation

PAY-142Checkout redesign

Product
Northwind
Team
Payments
Scope
payments/*ui/checkout/*
Claimed spend
£4,210
AIMS Evidence Check
  1. Match commits & files
  2. Check scope & branches
  3. Validate against allocation
  4. Flag exceptions
Observed work
  • 12 commits
  • 8 files touched
  • payments/*
  • ui/checkout/*
  • PRs & branches
Verdict
Match

Work produced matches the claimed story.

£4,210 ready to bill

Evidence trace for PAY-142

Real work that supports the claimed allocation.

Illustrative
  1. Commit a3f9c2e Add checkout payment flow 2 days ago
  2. File payments/stripe.ts Updated payment integration 2 days ago
  3. File ui/checkout/Confirm.tsx Update checkout UI 3 days ago
  4. Branch feature/checkout-redesign Feature branch for PAY-142 3 days ago
  5. PR #482 Merge checkout redesign 2 days ago
05 · A story, end to end

Make the cost of a feature legible

From authorised story to AI usage, every part of delivery has a place in the record.

  1. Authorised story

    Start with a clear budget, team, and sprint.

    • PAY-142
    • Northwind
    • Payments
    • Sprint 24
    • £5,000 budget

  2. AI usage

    Track which models and tokens were used.

    • GPT-4.1£2,540
    • Claude£1,670
    • 2.4M tokens

  3. Delivery context

    Understand where the work sits.

    Product
    Northwind
    Team
    Payments
    Sprint
    Sprint 24

  4. Feature cost

    See total spend for the feature in one place.

    Total spent £4,210 of £5,000 84% used
06 · For consultancies

AIMS changes consultancy economics

Bill the work you did, protect the margin you quoted, and run development against the client's own model accounts.

Example: a £50,000 fixed-price client project flows through AIMS, which tracks AI usage, attributes its cost, and organises it by feature. That separates your own project margin from the model costs billed to the client's own accounts.

Client project

Checkout redesign

Fixed-price engagement
Quoted value
£50,000
Team
Your consultancy
Timeline
Sprint 24
AI-assisted stories14 of 18 delivered
AIMS
  • Track AI usage
  • Attribute costs
  • Organise by feature
Your project economics
Project margin
Within quote
£21,600
  • Delivery cost £28,400
  • Margin £21,600
  • of £50,000 quoted
Client model costs
Billed to the client's accounts
On their tenancy
Client-owned model usage
  • Checkout UI
  • Payments API
  • Order emails

Their models, their token bill, their liability.

Client delivery

Chargeback that stands up to audit

Bill AI development at feature level, with a clear record of what was delivered and what it cost.

Fixed-price work

Protect the margin you quoted

See what an AI-assisted story actually costs and manage overruns with real usage data.

Client tenancy

Keep client costs with the client

Run development against their own model accounts and keep provider liability where it belongs.

07 · Why AIMS, not the alternatives

Better reporting is coming. Governance won’t come with it

They give you
Spend by team, user, or cost centreSee where the money went, after it has been spent.
AIMS gives you
Spend by feature, tied to your backlogUnderstand spend in the context of what you are building.
They give you
A tag that says money was spentManual tagging with no link to the work.
AIMS gives you
Spend tied to the story on your boardAttributed automatically from your development workflow.
They give you
A bill you reconcile after the factReactive finance with delayed visibility.
AIMS gives you
A budget enforced before the spend happensGateway limits stop the overspend instead of reporting it.
They give you
Attribution inside one vendor's ecosystemSiloed data, limited to a single provider.
AIMS gives you
Attribution across your clients own providersOne view across OpenAI, Anthropic, Gemini, and more.
08 · For internal engineering

Spend by feature, not by person

Govern spend across teams, stop allowance leakage, and make model-cost optimisation a deliberate choice.

AI spend

  1. 01 Govern

    Governed spend

    See spend across offshore, outsourced, and in-house teams.

    Spend distribution
    • In-house
    • Outsourced
    • Offshore

  2. 02 Control

    Less leakage

    Bring unsanctioned work into view before it becomes a surprise.

    Policy enforcement
    • Authorised workflows
    • Unapproved usage
    • Under review

  3. 03 Optimise

    Model optimisation

    Choose capability where it earns its place.

    Model cost comparison Bar height shows relative cost

    An abstract comparison of four models by relative cost, with the one whose capability fits the task highlighted.

  4. 04 Insight

    Decision-ready insight

    Give engineering leaders a clear view of authorised AI work.

    Authorised vs unauthorised
    • Authorised
    • Unauthorised
09 · Developer clients

Governed access, without changing how developers work

Developers keep their workflow. AIMS runs in the editor, CLI, and Copilot with entitled models, zero key handling, and per-request attribution.

No long-lived provider keys on developer machines.

VS Code, GitHub Copilot, and the CLI all send requests through the AIMS gateway, which checks entitlement, issues short-lived credentials, and attributes each request before routing it to an approved model.

  1. VS Code extension

    Models appear directly in the developer's existing workflow.

    Synapx AIMS · GPT-4.1

  2. GitHub Copilot BYOK

    Governed model access without persistent provider keys.

    Key auto-rotated

  3. CLI

    One command routes Copilot CLI, Codex CLI, or Claude Code through AIMS.

    $ aims run
AIMS Governance layer
  • Entitled models Policy-based access
  • Short-lived credentials No static keys
  • Per-request attribution Track and allocate usage

Approved models

  • Claude Sonnet 4.5 Anthropic Via AIMS
  • Gemini 2.5 Pro Google Via AIMS
  • GPT-4.1 OpenAI Via AIMS
  • o4-mini OpenAI Via AIMS
10 · Built for governed development

One broker. Central control

Every request passes through one broker, so credentials, access, and revocation are controlled in one place.

Developers Your tools & workflows
  • VS Code
  • GitHub Copilot

Central broker

Policy Access Credentials

  • Anthropic
  • Azure OpenAI
  • Google
  • OpenAI
  • Other providers
11 · Simple to start

Start with governed spend. Scale with evidence

AIMS starts with the controls teams need first: short-lived access, feature-level attribution, project entitlements, and budgets that can be enforced before spend occurs.

    • Short-lived access
    • Feature-level attribution
    • Project entitlements
    • Budget enforcement
    • Spend tracked by feature
    • Policy adherence
    • Usage visibility
    • Audit-ready data
    • Onboard more teams
    • Expand projects
    • Add new models
    • Same controls as you grow
Governed AI spend
Feature spend
  • In-house
  • Outsourced
  • Offshore
Policy status
  • Authorised
  • Under review
  • Unapproved
Model usage
Spend trend
  • Authorised
  • Unauthorised
12 · Objections, answered

The hard questions, up front

Isn't this just timesheeting for AI?

It's timesheeting the developer never fills in. Allocation comes from the board and rides in the credential, so nothing is typed and nothing is remembered. Checking that claim against the diff is what we're building next.

Won't developers mis-charge to whichever story has budget left?

The story is pre-selected from the active branch and the budget is enforced per story, so mis-charging takes deliberate effort and shows up in someone else's numbers. The scope check we are building on top of this is designed to narrow the gap further.

AI is getting cheaper anyway, so why govern it?

Capability per pound is falling, but the frontier models you actually use are flat to more expensive while consumption rises. Your bill is going up.

Can't our model vendor do this natively?

Only per developer, and only inside their own walls. Cross-vendor, SDLC-native, consultancy-ready attribution on your client's own accounts is not their business.

Know what your AI built, and what it cost

See how AIMS turns AI development spend into evidence you can govern, explain, and bill.

Request early access
Cross-vendorSDLC-nativeClient-owned accounts