Interface LDAIClient

Retrieves and processes a single AI Config agent based on the provided key, LaunchDarkly context, and variables. This includes the model configuration and the customized instructions.

Example

const key = "research_agent";
const context = {...};
const variables = { topic: 'climate change' };
const agentConfig = await client.agentConfig(key, context, {
  enabled: true,
  model: { name: 'gpt-4' },
  provider: { name: 'openai' },
  instructions: 'You are a research assistant.',
}, variables);

if (agentConfig.enabled) {
  const tracker = agentConfig.createTracker();
  const researchResult = agentConfig.instructions; // Interpolated instructions
  tracker.trackSuccess();
}
Copy

Retrieves and processes multiple AI Config agents based on the provided agent configurations and LaunchDarkly context. This includes the model configuration and the customized instructions.

Example

const agentConfigsList = [
  {
    key: 'research_agent',
    defaultValue: {
      enabled: true,
      model: { name: 'gpt-4' },
      provider: { name: 'openai' },
      instructions: 'You are a research assistant.'
    },
    variables: { topic: 'climate change' }
  },
  {
    key: 'writing_agent',
    defaultValue: {
      enabled: true,
      model: { name: 'gpt-4' },
      provider: { name: 'openai' },
      instructions: 'You are a writing assistant.'
    },
    variables: { style: 'academic' }
  }
] as const;
const context = {...};

const configs = await client.agentConfigs(agentConfigsList, context);
if (configs["research_agent"].enabled) {
  const tracker = configs["research_agent"].createTracker();
  const researchResult = configs["research_agent"].instructions; // Interpolated instructions
  tracker.trackSuccess();
}
Copy

Fetches an agent graph configuration from LaunchDarkly and returns an AgentGraphDefinition.

When the graph is enabled the method validates that:

• The graph flag can be evaluated.
• A single root node is present.
• All nodes in the graph are reachable from the root (no disconnected nodes).
• Every referenced agent config can be fetched and is enabled.

If any validation check fails, the returned definition has enabled set to false with an empty node collection. When the logger level is DEBUG, a message describing the failure is emitted.

Example

const graph = await aiClient.agentGraph('my-agent-graph', context, { userName: 'Sandy' });
if (graph.enabled) {
  graph.traverse((node, ctx) => {
    // build your provider-specific node here
  });
}
Copy

Retrieves and processes a completion AI Config based on the provided key, LaunchDarkly context, and variables. This includes the model configuration and the customized messages.

Example

const key = "welcome_prompt";
const context = {...};
const variables = {username: 'john'};
const defaultValue = {
 enabled: true,
 model: { name: 'gpt-4' },
 provider: { name: 'openai' },
};

const completionConfig = await client.completionConfig(key, context, defaultValue, variables);
if (completionConfig.enabled) {
  const tracker = completionConfig.createTracker();
  // Use completionConfig.messages and completionConfig.model with your LLM,
  // then record metrics with tracker.trackSuccess(), tracker.trackTokens(), etc.
}
Copy

Creates and returns a new ManagedAgent instance for agent interactions. Evaluations are wired automatically and exposed on ManagedResult.evaluations.

Reconstructs an LDGraphTracker from a resumption token, preserving the original runId so events from a resumed session are correlated correctly.

Security note: The token encodes the flag variation key and version. Keep it server-side; do not expose it to untrusted clients.

Creates and returns a new Judge instance for AI evaluation.

Example

const judge = await client.createJudge(
  "relevance-judge",
  context,
  {
    enabled: true,
    model: { name: "gpt-4" },
    provider: { name: "openai" },
    evaluationMetricKey: '$ld:ai:judge:relevance',
    messages: [{ role: 'system', content: 'You are a relevance judge.' }]
  },
  { metric: "relevance" }
);

if (judge) {
  const result = await judge.evaluate("User question", "AI response");
  console.log('Relevance score:', result.evals.relevance?.score);
}
Copy

Creates and returns a new ManagedModel instance for LLM model interactions.

Example

const key = "customer_support_chat";
const context = {...};
const defaultValue = {
  enabled: true,
  model: { name: "gpt-4" },
  provider: { name: "openai" },
  messages: [
    { role: "system", content: "You are a helpful customer support agent." }
  ]
};
const variables = { customerName: 'John' };

const model = await client.createModel(key, context, defaultValue, variables);
if (model) {
  const result = await model.run("I need help with my order");
  console.log(result.content);
}
Copy

Reconstructs an AIConfigTracker from a resumption token string previously obtained from a tracker's resumptionToken property. Use this to associate deferred events (such as user feedback) with the original tracker's runId.

Retrieves and processes a Judge AI Config based on the provided key, LaunchDarkly context, and variables. This includes the model configuration and the customized messages for evaluation.

Example

const judgeConf = await client.judgeConfig(key, context, {
  enabled: true,
  model: { name: 'gpt-4' },
  provider: { name: 'openai' },
  evaluationMetricKey: '$ld:ai:judge:relevance',
  messages: [{ role: 'system', content: 'You are a relevance judge.' }]
}, variables);

if (judgeConf.enabled) {
  const tracker = judgeConf.createTracker();
  // Use judgeConf.messages and judgeConf.model with your LLM,
  // then record metrics with tracker.trackSuccess(), tracker.trackJudgeResult(), etc.
}
Copy