BlogBlogConcentrate attributes AI coding costs to individual employees and teams.
Concentrate.ai
A single AI invoice tells finance what the company spent. It does not tell an engineering manager which employee, coding tool or team created the bill.
To track Claude Code and Codex API spend by employee, give each employee a distinct gateway credential, send supported API requests through that gateway, and map those credentials to the employee’s team and department. Keep subscription charges and requests that bypass the gateway in a separate ledger.
Concentrate provides a shared point for model access and spend tracking across providers. The useful outcome is an attributable API bill: who spent the money, which models they used and which budget owns the charge.
Start with three buckets: fixed subscription fees, API requests routed through Concentrate, and API requests billed elsewhere. A gateway cannot report a subscription charge or a direct provider request it never receives.
Check each coding tool’s authentication path before rollout. A developer signed into a subscription account may be using a different billing path from a developer using an API credential. There is no universal employee-count threshold at which subscriptions stop being economical.
A shared engineering key makes the total visible but hides individual usage. Issue a separate credential for each employee and record its owner, team, department and approved use. Keep service accounts separate from human users.
For example, an employee in the Payments team might own one credential for interactive coding. A continuous integration agent should have another credential owned by the Payments service account. That separation prevents an overnight agent run from appearing to be the employee’s interactive work.
Use the gateway’s supported identity and key-management controls. A free-form client label is useful metadata, but it should not be treated as a trusted employee identity on its own. Revoke or rotate credentials when ownership changes.
Configure the tool with the supported gateway endpoint, credential and model identifier. Concentrate’s API documentation provides the connection details. Confirm compatibility with the specific tool, model and provider route before rolling it out to the team.
Run a small test request, then check that its cost appears against the intended credential and model. Repeat for each tool and authentication method. If requests do not appear, resolve the billing path before relying on the report for employee attribution.
A spend report explains what happened. An alert tells someone to investigate. An enforced limit changes whether additional requests can proceed. Buying a dashboard does not automatically establish a daily spending policy.
Write down the monthly allocation, the desired daily guardrail, the alert recipient and the exception process. Then confirm which controls and reset periods are supported at the organization, team and key levels. Do not assume a monthly allowance prevents a large burst within one afternoon.
For example, a team with a $20,000 monthly allocation might choose an employee alert threshold of $50 a day. That is an illustrative policy, not a recommended default or a promise that every limit type is available at every scope. Set thresholds around actual workloads, and account for reporting delay and requests already in flight.
Concentrate’s spend tracking and alerts provide the reporting and notification pieces. Assign someone to receive and act on those alerts.
Employees can evaluate open-weight models for suitable work while retaining access to frontier models for harder tasks. Keep both paths attributable to the same employee and budget owner.
Compare completed-task cost, retries, cache behavior, latency and code quality. A cheaper token price does not guarantee a cheaper completed task. Use a harness and model combination that supports the workflow; do not assume every model works in every coding tool.
Roll employee credentials into stable team and department mappings. Review unassigned usage and shared credentials before drawing conclusions from the totals. Preserve subscription charges separately so finance can reconcile the full bill.
Try Concentrate to centralize supported API model access and attribute spend across your employees and teams. Start with one team, verify its requests and alerts, and expand once the accounting matches the usage.
LLM Gateway
Change models, providers, add fallbacks, and more by running all AI usage through one platform.