Postman becomes Git-native and gears up for AI agents in API development
Postman 12 introduces YAML-based Collections, Agent Mode, and a central API catalog – geared towards agent-driven development.
(Image: Pavel Ignatov/Shutterstock.com)
API toolmaker Postman has unveiled a fundamentally revised version of its development platform. The update makes the application completely Git-native and introduces new functions with Agent Mode and a central API Catalog, which aim at collaboration with autonomous AI agents.
According to the announcement in the Postman blog, APIs are increasingly becoming the critical interface between agents and the real world. The central idea: In contrast to the previously deterministic approach, AI agents make probabilistic decisions at runtime about which APIs they call, when, and in what order. Faulty or unreliable interfaces could quickly have cascading effects in agent-controlled systems. The innovations in Postman are therefore intended to prepare development teams for increasingly agent-driven software development.
Git-native architecture and new Collection format
Arguably the most profound restructuring concerns the way of working with version control. According to the manufacturer, the new Postman version is built from the ground up to be Git-native. Developers should be able to work in Postman on the same branch where they also write code – parallel to their IDE. The Git-native architecture also enables offline work.
(Image:Â avdyachenko/Shutterstock)
The online conference betterCode() API from iX and dpunkt.verlag will showcase modern API concepts over two days (May 12 and 21, 2026): protocols, routing, testing, etc. API security is also a topic, as is the new LLM interface Model Context Protocol (MCP).
An essential detail for everyday developer life: Postman introduces the new Collection 3.0 format, which uses YAML files instead of JSON blobs. Collections are split into individual YAML files. The files are thus intended to be readable and writable not only for AI agents but also easier for humans to compare and review. All Postman assets, including specs, flows, and local mock servers, are versioned along with the code.
Also new are code-based local mock servers that simulate API servers and can run both locally and in the CI pipeline. Postman expects more flexibility from this than from purely static mocking: mock servers are thus intended to move more to the center of development, for example when designing new APIs or stubbing dependencies.
Multi-protocol support in one Collection
Modern software systems rarely use only a single protocol, yet most tools treat each protocol separately. Postman now allows teams to organize HTTP, GraphQL, gRPC, MCP, MQTT, WebSockets, and AI requests in the same collection. Automation and validation across HTTP, GraphQL, and gRPC should be possible in the Collection Runner, with more protocols to follow. According to Postman, this results in system-wide testing that should reflect the actual end-to-end behavior of systems – without the coordination effort that arises when each component is validated in a different tool.
Revised CLI closes gap between local and CI
The Postman CLI should in the future be able to execute the same collections, tests, and mocks both locally and in CI pipelines, without workflows needing to be reconfigured for each environment. This should reduce CI-specific errors – for example, that gaps in test coverage only become visible after a commit – and standardize workflows.
Videos by heise
Agent Mode: AI as an active development partner
An AI function, which is intended to work across Postman and connected repositories, is now available under the name “Agent Mode.” According to the announcement, Agent Mode can edit, update, and create new Collections, tests, and mocks that follow the standards of the respective organization. Developers can use the AI via conversation, transfer complete workflows to it, or let it work directly on the codebase – for example, to fix errors, generate server stubs, or create client code. Postman assets can also be completely recreated by referencing existing code.
In addition, AI-assisted test generation should automatically create contract, load, unit, integration, and end-to-end tests for APIs. In the event of failed tests, for example in the Collection Runner, Agent Mode should be able to diagnose the cause and suggest a fix directly in the results – so that developers no longer have to painstakingly inspect requests, variables, and environments individually.
New API Catalog as central control level
One of the biggest problems in development organizations: There is no single place that answers fundamental questions about one's own APIs – which APIs exist, whether they are tested, whether they comply with internal standards, and how they perform in production. According to Postman, this information is spread across Git repos, CI dashboards, APM tools, wikis, and informal team knowledge.
The new API Catalog is intended to serve as an operational layer for API portfolio management and function as a “System of Record” that remains up-to-date because it is directly connected to the places where APIs are built, tested, and operated. Teams should be able to view their entire API landscape independently of the underlying infrastructure and across all environments.
The catalog integrates API governance, allowing central teams to enforce design rules, and offers analytics for measuring API health. Using Agent Mode, users should be able to query the catalog in natural language – for example: “Which APIs in production have no OpenAPI spec?” or “Which endpoints have a P95 latency over 500 ms in staging?” According to Postman, Agent Mode has access to the catalog's complete data model and can link governance, test, and runtime data in a single query.
Private API Network and new UI
Further innovations in Postman include the Private API Network, which has been updated on the publisher and consumer sides. Changes from Git now synchronize automatically with the network via the Postman CLI. The also revised UI now offers a unified workbench where Collections, Environments, Specs, Flows, and local mock servers can be organized together.
The Release Notes for Version 12 (currently 12.0.3) list all specific changes in detail.
(map)