Understand a repository, investigate React risks, compare code and review a component before bringing it into your project. SAGE turns that evidence into focused context for developers and AI coding agents.
Local analysis. CLI and MCP. No hosted code upload required.
Existing analysis and governance surfaces work from repository evidence. Their results help you choose where to inspect and what to verify before a change.
Build bounded dependency packages, inspect missing files and target conflicts, and review static adoption recommendations in a decision cockpit. AI task packs carry verification and rollback requirements.
Explore component adoption reviewUse symbol DNA and structural fingerprints for clone discovery, variation comparison and nanometric structural diff. Similar structure is review evidence, not proof of identical behavior.
Explore the structural modelMap projects, symbols, imports and routes. Inspect indexed source context and dependency relationships through the same architecture model.
Explore the architecturePost-Atlas Oracle proposes an architectural blueprint. Applicable policy and HITL govern enforcement; a guessed folder name is not enough.
How Oracle worksInspect React Compiler readiness, hook and render-risk signals, and Next.js server/client boundaries. Static findings keep their confidence and runtime-proof needs.
React capability evidenceReview static state/data flow in a project or focus the query on a file or symbol. Recorded transitions are evidence to inspect, not executed runtime behavior.
Agent inspection workflowRequest scoped context, change signals, dependency impact, static test-impact candidates and upstream cause traces instead of an undifferentiated code dump.
Connect a target-safe agentReview architecture drift, circular dependencies and unreferenced-code candidates with their evidence. A finding does not authorize deletion, a patch or a merge.
Read the evidence modelWatchdog refreshes bounded changed-source evidence and exposes active signals. Test-impact candidates, deferred proof and source drift stay explicit rather than becoming a false all-clear.
Follow the live focus loopBefore adopting code from a declared variation or companion, review its dependency closure and target fit. SAGE's Merge layer assembles static evidence and an actionable review plan; it does not blindly copy code or resolve Git conflicts.
Check Merge evidence and scopeSAGE is not another coding agent. It supplies repository context and static evidence to the developer or agent already doing the work.
Read the progressive HITL runbookStart with a real refactoring decision, then inspect the source and the evidence behind it. These cases demonstrate selected capabilities, not a repository-wide score.
In Bulletproof React's Next Pages and React/Vite applications, SAGE grouped matching profile components, mutation hooks and input schemas. Symbol search and source-bound inspection located the consumer; focused State Flow orientation exposed the hook's recorded references and unresolved calls.
Source review turns those candidates into a concrete checklist: retain user-data refresh, caller success feedback and pending/completion bindings. Auth and API integration still differ. This supports a sharing decision; it does not approve a merge.
Explore the profile-sharing decisionThis SAGE v1.4.0 case uses two actually indexed applications from a frozen checkout. The configured MAIN application was not acquired; the run's quality gate was FAIL. Those results remain in the full case, not a target-defect verdict. No target tests or components were executed. Merge had zero candidates, and static orientation did not resolve the complete state chain.
Inspect the selected evidence and limitsOn a frozen Zustand checkout, SAGE located createStore in src/vanilla.ts and returned src/index.ts, src/react.ts and src/traditional.ts as dependents. Each was checked against the source.
This SAGE v1.4.0 example demonstrates dependency and symbol orientation in the selected source scope. The full case retains unsuccessful checks and missing test coverage. No target tests or components were executed, and an empty candidate list does not prove no tests exist.
Read the scope, checks and unresolved resultsThe preserved Dub example separates a Next.js [domain] route from a DDD layer. Current static controls check generalized layer matching and policy applicability before recommending a refactor.
The case distinguishes the old v1.0.4 finding from current v1.4.0 static checks. It is not a fresh full Dub scan, a clean-repository verdict or a precision measurement.
Reproduce the static controlsFor the wider product boundary, see the claims/evidence matrix. Runtime-only behavior, equal-depth polyglot analysis and full SAST replacement are not current claims.
Acquire the public source and inspect prerequisites. The Quickstart explains local setup, analysis scope, outputs and connecting an agent.