A secret token used to authenticate your application with the Arcbeam platform. API keys are created in the Arcbeam dashboard and should be stored securely in environment variables.
A curated set of traces grouped together for team review and analysis. Collections allow you to organize interesting or problematic traces, add comments, and collaborate with stakeholders on improvements.
The Arcbeam client library that instruments your LLM application to send traces to the Arcbeam platform. Available for Python (LangChain/LangGraph) with additional framework support coming soon.
An external system connected to Arcbeam, such as a vector database (e.g., pgvector). Data sources enable Arcbeam to link traces with the actual documents retrieved during execution.
A collection of documents from a connected data source (vector database). Datasets represent your knowledge base and can be analyzed to understand document usage and quality.
A label applied to traces to distinguish between different deployment contexts (e.g., “development”, “staging”, “production”). Environments help organize and filter traces in the dashboard.
An individual step within a trace representing a single operation, such as a model call, retrieval, or tool execution. Traces are composed of multiple nodes organized in a parent-child hierarchy.
A complete record of a single LLM interaction, including all steps from input to output. Traces capture model calls, retrievals, tool executions, timing, costs, and errors.
An autonomous system that uses an LLM to reason about tasks, make decisions, and execute actions using tools. Agents can perform multi-step workflows without explicit programming for each step.
The maximum amount of text (measured in tokens) that an LLM can process in a single request, including both input and output. Different models have different context window sizes (e.g., 4K, 8K, 128K tokens).
A numerical vector representation of text that captures semantic meaning. Embeddings enable similarity search in vector databases and are fundamental to RAG systems.
When an LLM generates information that is factually incorrect, fabricated, or not grounded in its training data or provided context. A key challenge in production LLM applications.
An open-source framework for building applications with LLMs. Provides abstractions for chains, agents, memory, and integrations with various LLM providers and tools.
A library built on LangChain for creating stateful, multi-step agent workflows using graph-based orchestration. Enables complex agent behaviors with cycles and conditional logic.
A technique that combines information retrieval with LLM generation. The system retrieves relevant documents from a knowledge base and includes them in the prompt to ground the LLM’s response in factual information.
Instructions provided to an LLM that define its behavior, role, and constraints for a conversation. System prompts guide the model’s responses and are typically invisible to end users.
A parameter controlling the randomness of LLM outputs. Lower values (e.g., 0.0-0.3) produce more focused and deterministic responses, while higher values (e.g., 0.7-1.0) increase creativity and variation.
The basic unit of text processed by LLMs. Tokens can be words, parts of words, or punctuation. Most LLMs charge based on token usage, and context windows are measured in tokens.
When an LLM decides to invoke an external function or API to retrieve information or perform an action. Tools extend LLM capabilities beyond text generation.
A specialized database optimized for storing and searching embeddings. Vector databases enable semantic search by finding documents with similar meaning rather than exact keyword matches. Examples include pgvector, Pinecone, and Weaviate.
The time elapsed between sending a request and receiving a response. In LLM applications, latency includes model processing time, retrieval time, and network overhead.
An open-source observability framework for collecting traces, metrics, and logs from applications. Arcbeam uses OpenTelemetry as the underlying protocol for trace collection.
A single unit of work within a distributed trace, representing an operation with a start time, duration, and metadata. In OpenTelemetry, spans are the building blocks of traces (similar to Arcbeam nodes).
A segment of a larger document split into smaller pieces for embedding and storage in a vector database. Chunking strategies balance context preservation with retrieval precision.
The ability to track data from its source through transformations to its final use. In Arcbeam, data lineage connects traces to the specific documents retrieved from vector databases.
Additional information attached to documents, traces, or nodes. Metadata can include timestamps, user IDs, document sources, or custom attributes for filtering and analysis.
A numerical value indicating the relevance of a retrieved document to a query. Higher scores suggest stronger semantic similarity. Also called similarity score or relevance score.
The total number of tokens consumed by an LLM interaction, including both input and output tokens. Used to calculate costs and track resource consumption.
