Set up AI review summaries
Connect an Anthropic or OpenAI key, pick a default model, and choose per screen or live coding session whether summaries run automatically.
AI review summaries run on your organization's own Anthropic or OpenAI key, the same connection the AI agents use. Anthropic or OpenAI bills your account for it. Summaries are part of the Pro and Enterprise plans.
Connect a key
An owner connects an Anthropic or OpenAI key under Integrations → AI assistants. See Connect your Anthropic and OpenAI keys. Without a key, the summary card on the review page says so and links to the place to add one.
Pick a default model
Open Integrations → AI review summary. Under Default model, an owner picks the model summaries run on:
- Claude Opus 5.5, Claude Sonnet 5.5 or Claude Haiku 5.5 with an Anthropic key
- GPT-6 Astra, GPT-6.1 Sol or GPT-6 Luna with an OpenAI key
Leave it on Automatic and Kendor uses the first model of a vendor you have a key for. A reviewer can still pick another model each time they run a summary.
Run it automatically
In a screen's settings, under Options, switch on AI review summary. Each attempt is then summarised shortly after it's submitted or times out. For a live coding interview, the same switch in the session's settings summarises each session when it ends.
With the switch off, nothing runs on its own, but owners and editors can still run a summary by hand from any finished attempt's review page.
Add extra instructions
Below the switch, Extra summary instructions (up to 2,000 characters) tell the model what to focus on for this screen, for example "Focus on API design and error handling". They're added after Kendor's own prompt. They can shift the focus but not its rules: no scores, and evidence for every claim.
Limits and cost
Summary tokens count toward the organization's Monthly token cap under Integrations, alongside the candidates' AI agents. Once the cap is reached, new summaries are refused until the next month.
What candidates are told
When your organization can run summaries, the candidate's start page, and the lobby of a live coding session, tell them before they begin that your organization may use AI, on its own Anthropic or OpenAI account, to summarise their work for reviewers, that it gives no score, and that people make every decision.