Execution-flow aware
Understands behavior, not files
Nexus traces call paths, processes, and dependencies before touching code.
Private betaNEXUS AGENT / 01
More performance. Fewer tokens.
Nexus Agent builds GitNexus into the coding harness itself. Execution flows, blast radius, and cross-repo structure stay inside the agent loop—so models spend more of their context solving and less of it rediscovering the codebase.
1.85× solve rate vs bare modelFewer tokens + stepsGitNexus built into the loop
nexus — payments-api
$ nexus agent
What are we building?
Add idempotent retries to failed invoice capture.
Tracing invoice capture execution flow…
✓21 symbols · 4 services · 2 downstream consumers
✓Blast radius checked before edit
PlanEditVerify
Built on the graph
The harness is the advantage.
GitNexus does not sit beside the agent as another search tool. Its resolved graph stays in the loop through planning, editing, and verification.
Blast-radius checks
Sees what an edit can break
Every change starts with upstream impact and ends with a system-level verification pass.
Cross-repo context
Works across service boundaries
One graph connects contracts, symbols, and flows across all repositories in your workspace.
Agent loop
Terminal speed. System-level judgment.
Start in Nexus CLI or continue in web. Same graph, same sessions, same understanding of what your change touches.
01MapLoad relevant symbols, execution flows, and service contracts.
02PlanShow intended edits and blast radius before code changes.
03BuildWork from terminal or web with GitNexus context attached.
04VerifyRe-check affected flows, tests, and downstream behavior.
Benchmark proof
Same model. Better harness.
On DeepSWE, GitNexus raised the same model's solve rate from 37.0% to 68.4%, while using fewer tokens, fewer steps, and 51% less cost per solved task.
GitNexus-native agent loop
Nexus CLI + web agent
Execution-flow context
Cross-repo blast radius
DeepSWE · same model
68.4%
GitNexus solve rate
37.0%
bare-model solve rate
$0.88 vs $1.79 per solved task
Private beta access opens in small cohorts.