Swiggy opens food, Instamart and Dineout services to external AI agents

That setup is meant to let outside AI assistants carry out tasks on top of Swiggy’s systems. Rao gave one example: users could connect Swiggy’s MCP to a large language model and ask it to plan meals around a dietary schedule, using Swiggy’s infrastructure to turn those requests into actual transactions (Inc42).

Building for interchangeable models

Rao also described a broader effort to avoid overreliance on any single AI model. He said Swiggy had previously built a contact-centre agent around one model provider and found that moving away from it after capacity constraints took nearly a month, pushing the company to focus more on workflows and evaluation systems rather than tying products too closely to one model.

Swiggy now uses an LLM gateway to route tasks to different models and tests alternatives in live operations, while also using models across cloud providers. Rao said the company chooses among them based on availability, response time and cost (Inc42).

Restricting agent access

As it expands the role of AI agents, Swiggy is also separating them from core production systems. Rao said the company is deploying agents in separate production-like environments with restricted ingress and egress rules instead of placing them alongside core services, part of a wider effort to limit access and work through questions around identities, permissions and trust (Inc42).

Internally, Swiggy is already using AI in delivery partner onboarding. Rao said the company’s in-house assistant guides new riders through the process in their local language and is improving rider net promoter scores and the onboarding funnel, though he did not share figures (Inc42).