Agentic Adapters

Agentic Adapters

ArtemisKit provides specialized adapters for testing agentic AI systems built with LangChain and DeepAgents.

Tip

Agentic adapters capture execution traces, tool calls, and inter-agent communication for comprehensive testing.

Available Adapters

LangChain Adapter

Test chains, agents, and runnables built with LangChain.js. Supports LCEL, ReAct agents, RAG chains, and more.

Learn more →

DeepAgents Adapter

Test multi-agent systems built with DeepAgents. Supports sequential, parallel, and hierarchical workflows.

Learn more →

Installation

LangChain

bun add @artemiskit/adapter-langchain
# or
npm install @artemiskit/adapter-langchain

DeepAgents

bun add @artemiskit/adapter-deepagents
# or
npm install @artemiskit/adapter-deepagents

Quick Comparison

Feature	LangChain	DeepAgents
System Type	Chains, Agents, Runnables	Multi-agent teams, hierarchies
Execution	Single-agent or chain flow	Multi-agent collaboration
Tool Tracking	Intermediate steps	Cross-agent tool usage
Communication	Chain input/output	Inter-agent messages
Streaming	Via stream() method	Via stream() method

Common Patterns

Testing Agent Tool Usage

Both adapters capture tool call information for verification:

import { createLangChainAdapter } from '@artemiskit/adapter-langchain';

const adapter = createLangChainAdapter(myAgent, {
  name: 'tool-agent',
  captureIntermediateSteps: true,
});

const result = await adapter.generate({
  prompt: 'Search for weather in New York',
});

// Verify tools were used correctly
const metadata = result.raw.metadata;
expect(metadata.toolsUsed).toContain('web_search');
expect(metadata.totalToolCalls).toBeGreaterThan(0);

Testing Multi-Agent Workflows

import { createDeepAgentsAdapter } from '@artemiskit/adapter-deepagents';

const adapter = createDeepAgentsAdapter(myTeam, {
  name: 'content-team',
  captureTraces: true,
  captureMessages: true,
});

const result = await adapter.generate({
  prompt: 'Research and write an article',
});

// Verify agent collaboration
const metadata = result.raw.metadata;
expect(metadata.agentsInvolved).toEqual(['researcher', 'writer']);
expect(metadata.totalMessages).toBeGreaterThan(0);

Using with ArtemisKit Scenarios

Create YAML scenarios for agentic testing:

agentic-test.yaml

name: agent-quality-test
description: Test agent response quality

cases:
  - id: tool-usage
    prompt: "Calculate 25 * 4 using the calculator"
    expected:
      type: contains
      values: ["100"]
    tags: [tool-test]

  - id: multi-step
    prompt: "Search for today's weather and summarize it"
    expected:
      type: llm_grader
      criteria: "Response includes current weather information"
      minScore: 0.8
    tags: [integration]

Run with ArtemisKit:

import { ArtemisKit } from '@artemiskit/sdk';
import { createLangChainAdapter } from '@artemiskit/adapter-langchain';

const adapter = createLangChainAdapter(myAgent);
const kit = new ArtemisKit({
  adapter,
  project: 'agent-testing',
});

const results = await kit.run({
  scenario: './agentic-test.yaml',
});

console.log(`Pass rate: ${results.manifest.metrics.pass_rate * 100}%`);

Execution Metadata

Both adapters provide rich execution metadata:

Field	Description
agentsInvolved	List of agents that participated
totalAgentCalls	Number of agent invocations
totalToolCalls	Total tool/function calls
toolsUsed	Unique tools that were used
traces	Full execution trace (if enabled)
messages	Inter-agent messages (if enabled)
executionTimeMs	Total execution time

Best Practices

1. Enable tracing in development — Use captureTraces: true for debugging
2. Set appropriate timeouts — Multi-agent systems can take longer
3. Test tool usage — Verify agents use tools correctly
4. Test failure paths — Ensure graceful handling of tool errors
5. Monitor costs — Multi-agent systems can make many LLM calls

See Also

• LangChain Adapter — Full LangChain documentation
• DeepAgents Adapter — Full DeepAgents documentation
• SDK Overview — ArtemisKit SDK documentation
• Test Matchers — Jest/Vitest matchers