Manage AI API access without sharing the original keys.
Developers and teams using AI provider APIs across scripts, services and individual workstations.
Flume is a proxy between an application and its AI providers. Instead of copying a provider’s main API key into every script or teammate’s environment, a team gives each consumer its own token. We built the product around controlling that access and seeing which consumer generates the requests, token usage and estimated cost.
The provider key is stored encrypted on the server. Applications use issued proxy tokens instead, each tied to a provider. A team can give its development scripts, production service and individual users separate tokens. Revoking one consumer does not require distributing a new provider key to everybody else.
Tokens can carry an expiry, a model allowlist, daily request limits and a spend cap. The proxy checks these rules before forwarding requests. That makes access something the team can adjust per service or person, rather than an all-or-nothing permission attached to the original provider key.
The dashboard breaks activity down by token and provider, including request counts, latency and available token-usage data. Model pricing tables produce cost estimates, including relevant cache rates. These figures help investigate an unexpected bill or compare workloads; they are estimates rather than a replacement for the provider’s invoice.
Supported integrations retain the provider’s request paths, bodies and response shapes, with clients pointing at the proxy URL. Streaming responses pass through to the application. The proxy records usage information for metering rather than building a library of users’ prompts and generated responses.
Add the provider key to Flume and choose the account or service it represents.
Create a separate token for a consumer and set the limits or expiry it needs.
Point the client at the proxy, then inspect usage and revoke or adjust access as the workload changes.