Recommendations you can apply — not a PDF you file away.
After every simulation, a hybrid rule-engine and AI pass analyzes your canvas, simulation results, and cost data. It surfaces bottlenecks, service swaps, configuration changes, and scaling fixes — each with severity, rationale, and a one-click Apply button that updates the canvas itself.
Lambda / api-worker — Concurrency exceeded at 1,240 RPS. Add SQS buffer between ALB and worker pool.
DynamoDB / orders — On-demand is 3.1× provisioned cost at this profile. Est. saving: $690/mo.
CloudFront — Adding a CDN in front of S3 media cuts origin egress by ~82%.
Rules find the known failures. AI finds the rest.
A deterministic rule engine catches the well-understood anti-patterns — throttling limits, missing buffers, capacity misconfiguration. An optional AI pass reasons about your specific topology on top of it.
Recommendations are generated from the run you just executed — your traffic pattern, your peak RPS, your cost breakdown. Not generic best practices.
Deterministic rules catch throttling, queue overflow, and capacity issues. An optional Claude pass suggests architecture pattern changes and new services.
Every finding is ranked — warning, advisory, info — with a written explanation of the problem and the fix, so you can defend the decision in review.
Service swaps and configuration changes with the estimated monthly saving attached — on-demand vs provisioned, caching layers, right-sizing.
Cold-start mitigation, CDN placement, and connection pooling suggestions targeted at your P95/P99 under simulated load.
Missing DLQs, unbuffered fan-outs, and single points of failure surfaced before they become incidents.
The fix is a button, not a ticket.
Each recommendation carries an Apply action. Clicking it patches the node configuration or adds new services — an SQS buffer, a CloudFront CDN — and wires the edges automatically.
Simulate
Run your traffic pattern. The engine records per-node load, latency, errors, and cost.
Review findings
Recommendations arrive ranked by severity with rationale and estimated impact.
Apply
One click patches config or adds services and connections directly on the canvas.
Re-simulate
Validate the change against the same traffic. Both runs stay in execution history.
Optimization starts before you press Simulate.
The canvas itself flags problems while you design — you don't have to wait for a run to catch an anti-pattern.
Per-node warnings and suggestions — anti-patterns and missing configuration — shown right in the node configuration panel.
Invalid source-to-target pairs are blocked before they can be created, with error toasts that suggest the correct wiring.
Built-in guidance for each service explaining how it behaves at scale, common pitfalls, and distilled best practice.
Coming next on the roadmap: an optimisation dashboard tracking cumulative savings across all your architectures.
Grounded in your run
Findings come from your simulation data — your pattern, your peak, your cost — not a generic checklist.
Actionable by design
Every recommendation names the service, the problem, and the fix — and can be applied in one click.
Validated by re-run
Apply, re-simulate, compare. Execution history keeps the before-and-after as evidence.
Find the $690/mo
hiding in your design.
20 recommendations/month free · No cloud account required