What is CoreValue?

CoreValue is an AI gateway with spend control built in. Your applications send their AI requests through it; it checks each one against your budgets before it reaches the provider, records what it cost, and attributes that cost to the team, customer and feature it was for.

The problem

AI providers give you usage dashboards and spend limits, but they are organised around your account (projects, workspaces and keys), not around your customers, teams and features. Each provider has its own prices, reports and limits, and a limit stops all the traffic that shares it.

You can tag requests and add up costs in your own code. But holding each customer and team to a budget, across every provider, means building and maintaining that yourself.

CoreValue puts attribution and budgets in the request path. Every request is priced and attributed to its customer, team, member, key and feature. Before it reaches the model, it is checked against the budgets that apply. A key cap, member budget or customer budget refuses it; a team or organisation budget can move it to a cheaper model instead. Each check first reserves a worst-case estimate, so concurrent requests cannot all slip past the same budget.

How it works

  Your app or agent  ──►  CoreValue gateway  ──►  OpenAI · Anthropic · Google · more
                          │  checks budgets and policy before the provider
                          │  prices the response, settles budgets
                          ▼
                       Console: spend · budgets · alerts · reports

You change two lines: the base URL and the key.

# Before
client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

# After
client = OpenAI(
    base_url="https://gateway.corevalue.dev/v1",
    api_key=os.environ["COVA_API_KEY"],  # sk-cova-…
)

Your provider key moves into CoreValue, and the provider keeps billing you as it does today. See Architecture for the full path of a request.

What you get

Measure the cost

• Every request with its model, tokens, latency and cost. → Cost tracking
• Spend by team, member, customer, feature and model. → Cost attribution
• Multi-step agent runs followed from first request to last. → Runs

Control the spend

• Monthly budgets for the organisation, a team, a member, a customer or a key, checked before each request. → Budgets
• Refuse, downgrade to a cheaper model, or alert when a budget runs out.
• Rate limits, blocked models, key scopes and PII guardrails. → Access control
• Email alerts before a budget runs out. → Budget alerts

Prove the value

• Chargeback to teams for finance. → Chargeback
• Cost and margin per customer. → Unit economics
• An executive summary and scheduled reports. → Reports

Trust

• Your provider keys stay out of your code and are never sent to your applications.
• CoreValue keys are stored as hashes and cannot be read back.
• Roles decide who sees what in the console, and changes are recorded in the audit log.
• What is stored, and for how long, is set out in Data and privacy.

Get started

API reference and SDKs

• Authentication: how requests carry a CoreValue key.
• Headers: every Cova-* header you can send or receive.
• Errors: what each refusal looks like.
• CoreValue SDK: TypeScript and Python helpers for headers and cost.
• Integrations: LangChain, Vercel AI SDK, LlamaIndex and more.

Tools and agents can read these docs at /llms.txt.