Skip to main content

Overview

Start here to understand the fundamental concepts in Arcane. Each section provides a brief overview and links to detailed documentation.


Traces, Spans & Conversations

How execution data and journeys are captured and replayed. Traces are end-to-end executions; spans are the timed steps inside them. Conversations are groups of traces that share one or more common attribute identifiers (e.g. conversation id), configured in the Conversation tab—Arcane stitches them into a replayable journey. Events are timestamped records attached to spans.

Learn more →


Datasources

The trace backends you connect to Arcane—Tempo, Jaeger, ClickHouse, or Custom API. Datasources expose traces in OpenTelemetry format so Arcane can read and display them.

Learn more →


Entities

Human-friendly labels for spans that map your business entities (models, tools, agents, guardrails, etc.) to telemetry data. Entities enable detection, highlighting, message matching, badges, and filtering—making traces readable and actionable.

Learn more →


Datasets

Collections of items curated for evaluations and experiments. Datasets can come from traces or be imported from your own sources. They provide stable, repeatable inputs for scoring and A/B comparisons.

Learn more →


Prompts & Models

The prompt versions and model configurations tied to LLM generations. Prompts are versioned templates; models are registered AI providers (OpenAI, Anthropic, etc.) used in prompts, scores, experiments, and evaluations.

Learn more →


Experiments & Evaluations

How you test and measure AI quality. Experiments run prompts against datasets to generate outputs. Evaluations measure those outputs (or dataset rows) using scores to assess quality. Together they let you compare prompt versions, measure performance, and track regressions.

Learn more →


Annotations Queues

Work queues for human review and labeling. Use annotation queues to review traces or conversations, answer questions, and create labeled data for training or evaluation.

Learn more →


Organisations & Projects

How data is organized and shared. Organisations are top-level containers where teams share datasources, entities, and models. Projects are workspaces that group traces, conversations, evaluations, and experiments.

Learn more →


  • Configure First — Set up your instance, organization, and projects
  • Operate — Use traces, conversations, evaluations, and experiments