Your AI is your edge. Make it better than your competitors’.

Hobby

$99/month

• Generous usage A run is one full test of your app: simulated users hold realistic conversations with it on your machine, and every reply is checked. Runs and reruns are included, not billed individually. Usage depends on workload size and complexity. Fair-use limits apply.

Agent workflow

• Coding-agent integration Your coding agent can invoke Cortad after a change and receive run results showing regressions, improvements, and the steps and scenarios behind them.
• Adaptive runs Multi-turn simulated conversations against your actual application and its entire harness, every layer as it really runs: prompts, tools, retrieval, memory, routing, queues, database and auth. Scenarios come from your application’s context. Checks evaluate each reply and feed back into the simulated user, which adapts subsequent messages to probe the weaknesses it encounters.
• 100 checks per message Evaluate each message against 100 contextual behavioral checks derived from your prompts, code, and intended outcomes. Checks are customizable, and the same checks can evaluate both runs and connected production messages.
• Custom checks & metrics Define your own checks and the metrics they contribute to, including instruction-following, hallucination, efficiency, and application-specific outcomes. Inspect results at the message, conversation, and run level.

Analysis

• Real-time dashboard View production activity, behavioral metrics, and the conversations behind them as messages arrive.
• Metric grouping & segmentation Split any metric by group. An overall number can look healthy while first-time users struggle, and grouping shows the difference.
• Surfaces Each place your app’s AI answers, such as a chat, a tutor or a support agent, found in your code. Results are split by surface, so you can see which one needs work.

Production

• Production logging Connect your production messages, and every one is read with the same checks as your runs.
• Production-informed runs Feed observed production behavior and failures into future scenarios and simulated-user behavior, using the same checks in production and during runs.

Advanced capabilities

• Anomaly detection Flags a behavioral metric that shifts outside its usual range, as messages arrive. Not included
• Behavior investigations An investigation into every anomaly: the cause, the users affected, and a suggested fix. Not included
• Live user segmentation Segments form from how users actually behave, such as first-time users, power users or frustrated users. Watch each one move in real time. Not included
• Segment experiments Push prompts, models or features to any segment. Experiments are measured on behavior, with confidence intervals. Not included

Allowances

Analyzed production messages One production reply counts as one analyzed message, however many checks evaluate it.: 100k/month

History: 30 days

Team members: Unlimited