Skip to main content

Description

The Ellipsis SDK enables developers to collect and export telemetry from AI agents and LLM-powered applications to the E1 observability platform. Use the SDK to monitor supported agent executions, LLM invocations, tool calls, traces, latency and token usage through framework instrumentation.

Package

@ellipsis1/sdk on npm

View the published package, release information and installation instructions.
The published package listing identifies version 0.1.0 and Node.js >=22. Check npm for the latest release before upgrading or pinning a version.

Runtime requirements

The package provides both ESM and CommonJS bundles.

Supported integrations

Framework compatibility

LLM provider compatibility

The SDK captures telemetry emitted through the LangChain framework. It is not currently tied to a single LLM provider and can work with different models supported by LangChain, including OpenAI, Google Gemini, Anthropic, OpenRouter and other compatible providers.
Provider-specific instrumentation outside the LangChain framework is not currently supported.

Initialization

Initialize the SDK once at the top level of your application, after loading the required configuration.
initEllipsis() accepts a configuration object and an optional logging level.
Supported logging levels:
  • debug
  • info
  • warn
  • error
  • silent
The default logging level is warn.

Configuration reference

The SDK accepts a Config object through initEllipsis().

Validation constraints

The SDK logging level is passed as the optional second argument to initEllipsis(); it is not listed as a field in the Config object.

Configuration example

Using environment variables

Create a .env file:
Load the environment variables before initializing the SDK:
Do not commit .env files or expose your agent API key in public repositories. Use your deployment environment’s secret-management facilities in production.

Providing values directly

You can also provide configuration values directly:
Replace the placeholder values with your actual ingestion URL and agent API key. For production applications, environment variables or a secrets manager are preferable to hardcoded credentials.

LangChain instrumentation

Create the LangChain handler from the initialized SDK instance:
Pass it to the agent invocation using LangChain’s callbacks option:
The handler captures supported framework events, normalize them across different model providers and exports telemetry to E1.

What the SDK collects

The SDK organizes captured execution events into traces and hierarchical spans. The E1 platform can present:
  • Trace and span hierarchies.
  • Agent and chain executions.
  • LLM provider and model information.
  • Token consumption.
  • Prices per run.
  • Execution duration and latency.
  • Tool calls and associated operations.
  • LLM reasoning behind each step.
  • Errors and failed executions.
The SDK supports multi-agent and multi-model projects.

How it works

The SDK acts as an instrumentation layer between your application and the E1 observability platform. Its architecture consists of four main components:
  1. Instrumentation adapters connect to supported frameworks and capture execution events.
  2. Trace and span management organizes events into traces with parent-child relationships.
  3. Telemetry processing aggregates execution information, including duration, token usage and LLM metrics.
  4. Telemetry exporter batches and transmits collected telemetry to E1.
The SDK is designed to minimize its impact on application execution. Telemetry is exported using batching and retry mechanisms for transient network failures.

Module compatibility

The SDK supports both Node.js module systems:
  • ES Modules (ESM)
  • CommonJS (CJS)
The package provides dedicated ESM and CommonJS bundles.