Customer-owned Tavily API keys
Product manager at Cohere on enterprise AI search infrastructure and deep research agents
This setup turns web search from a bundled feature into customer owned infrastructure. Because North runs inside the customer’s own environment, the customer plugs in its own Tavily key, pays Tavily directly, and carries its own usage bill. That keeps Cohere out of the pass through search spend, fits private deployment requirements, and makes cost scale with each enterprise’s actual employee usage instead of forcing Cohere to subsidize heavy search traffic.
-
Tavily prices on a credit basis, not a large flat platform fee. Basic search costs 1 credit, advanced search costs 2, and pay as you go pricing is $0.008 per credit, with volume plans lowering the per credit cost. For a large enterprise, that usually makes search a small variable line item relative to the value of an internal AI assistant.
-
The key reason each customer needs its own key is deployment architecture. North can run in a customer VPC or on premises, and Cohere states that prompts, outputs, and models stay inside the customer environment. In practice, the external search connector has to be authenticated and billed from that same environment.
-
This also changes vendor power. Tavily gets a direct billing relationship with each enterprise, while Cohere keeps North focused on orchestration, model quality, and workflow integration. It is similar to broader AI search infrastructure, where providers like Tavily, Exa, and Parallel are often slotted in as interchangeable retrieval layers rather than deeply embedded product suites.
Going forward, enterprise AI stacks are likely to look more modular, with private assistant platforms owning workflow and governance while search vendors monetize usage directly. That favors vendors with simple per query economics and easy swap in integration, and it pushes application companies like Cohere to win on deployment trust and product experience rather than resale margin on search.