Skip to main content
Jina combines connected context with a harness that plans, runs, and follows up on production scenarios. Together, they help Jina identify more critical issues and provide a stronger merge gate.

​
Context graph

The context graph connects code and infrastructure so Jina can trace a change beyond the diff. In the demonstration below, a checkout change touches shared rounding code also used by refunds. A related incident gives the agents more context for investigating that path.
Checkout and refunds share money/round.ts; the Refunds feature connects to incident INC-71.

A checkout change connects to shared rounding code, refunds, and a past incident. From Jina's context-graph demonstration.

View the context-graph post.

​
Investigation harness

For a PR that changes checkout sessions, the planner sends investigators to test payment capture, sessions, cache behavior, webhooks, and retries in parallel. Their evidence guides the planner’s next round, within the configured depth limit.
In this checkout example, a retry captures a payment twice. The planner follows that finding into refunds and capture races, while a stale-cart finding leads to another cache investigation. Each round explores what the previous one exposed.
Checkout example: the planner sends parallel investigators to test payment capture, sessions, cache, webhooks, and retries. Round two follows findings into refunds, capture races, and wishlist caching, and checks rollout flags from PR context.

Example: a checkout-session PR leads from payment and cache findings to deeper investigations in round two.

View the harness post. Context reveals where to investigate. Execution tests what happens. Iteration follows the evidence deeper. Jina validates the findings and uses them to support the merge decision.