Skip to main content

Documentation Index

Fetch the complete documentation index at: https://trygradient.ai/docs/llms.txt

Use this file to discover all available pages before exploring further.

Candidates & Sessions

Candidates

A candidate is a person being evaluated. Candidates are identified by email and scoped to your organization. The same person can take multiple assessments, and each attempt creates a separate session. Candidates are not admin users - they don’t log into the admin portal. Instead, they receive a unique link that drops them directly into the assessment workspace.

Sessions

A session represents one candidate’s attempt at one assessment. It tracks everything: when they started, what they configured, every message they exchanged with the AI, and the final deliverable they submitted.

Session lifecycle

invited → setup → in_progress → submitted → scoring → scored → reviewed
StatusWhat’s happening
invitedCandidate received the link but hasn’t opened it yet
setupCandidate is configuring their workspace (connecting data sources, creating skills, setting memory)
in_progressCandidate is working on the task with AI assistance
submittedCandidate submitted their deliverable (or time expired)
scoringThe automated scoring engine is evaluating the submission
scoredScores are available for admin review
reviewedAn admin has reviewed and optionally adjusted the scores

What happens during a session

1

Setup phase

The candidate configures their AI workspace: connecting data sources, uploading or creating skills, writing memory entries, and editing their agents.md configuration. This phase has its own timer.
2

Task phase

The candidate works with the AI assistant to complete the task brief. They can chat, run tool calls (like creating slides), search connected data sources, and iterate on their deliverable. A countdown timer is visible.
3

Submission

The candidate submits their deliverable, or it auto-submits when time runs out. At submission, Gradient snapshots the AI configuration (skills, memory, agents.md, connected data sources) for scoring.

Session events

Every action a candidate takes during a session is recorded as a timestamped event. Events power session replay in the admin review interface and are used by the scoring engine to evaluate AI usage patterns. There are 20+ event types covering data source browsing, skill creation, AI interactions, deliverable edits, and more. See the Sessions API for the full list.