I am excited to announce the launch of the Postman Context Graph API! The Context Graph API lets an agent see your API ecosystem as a connected system. You do not need to replace your model, agent, or IDE. You give the tools you already use access to the... graph, and they can reason across services, APIs, deployments, runtime calls, and external systems that were previously invisible to them.show more

With the Context Graph, your agent is no longer limited to whatever code you put in front of it. In our benchmark, we asked each model to identify every service across a large public software ecosystem that exposes HTTP endpoints. The model could search the repositories in its local checkout, but that checkout represented only 129 of the 498 services the Context Graph surfaced. The other services lived in repositories outside the checkout, existed without their own repository, or were separate components buried inside larger repos. No amount of better searching could find all of them because much of the code was never available to search in the first place. You cannot grep what was never in the checkout.

We tested three different models across seven scored prompts, and the Context Graph improved accuracy in 18 of the 21 model-prompt pairs. It gave each model a much better place to start, so instead of searching the entire estate from scratch, it could focus on verifying the relationships that actually mattered.

You can really see the difference when you ask who calls an API before making a breaking change. On its own, Opus found 9 of the 13 real callers. With the Context Graph, it found 12. Its accuracy jumped from 0.777 to 0.943. It found more of the right callers and guessed less.

The model also did less work to get there. Across the benchmark, Opus cost 27% less with the Context Graph and Astra cost 20% less. On the HTTP-endpoint question, Opus went from $37 per run to $10, used about a quarter of the input tokens, and made half as many tool calls. Once the graph pointed it in the right direction, it did not need to burn time and tokens rediscovering the ecosystem one search at a time.

Checkout the Context Graph API here along with instructions to integrate your agents:

This looks dope, heard it runs on Daytona tho. If they cause downtime lmk. We'll provide 50k credits and get you back online then.