LangChain Integration

Use Cognipeer Console as a LangChain chat model, and capture LangChain runs as Cognipeer tracing sessions. These helpers are exported from the package root.

import {
  CognipeerLangChainChatModel,
  CognipeerTracingCallbackHandler,
  createCognipeerAgentTracing,
  createCognipeerTracingMiddleware,
} from '@cognipeer/console-sdk';

Requires @langchain/core as a peer dependency.

Chat model

CognipeerLangChainChatModel is a drop-in BaseChatModel that proxies to Console chat completions, so it works anywhere a LangChain chat model does (chains, agents, LCEL).

import { CognipeerLangChainChatModel } from '@cognipeer/console-sdk';
import { HumanMessage } from '@langchain/core/messages';

const model = new CognipeerLangChainChatModel({
  model: 'gpt-4o-mini',
  client: { apiKey: process.env.COGNIPEER_API_KEY! },
  temperature: 0.2,
});

const res = await model.invoke([new HumanMessage('Summarize LangChain in one line.')]);
console.log(res.content);

Constructor params (CognipeerLangChainChatModelParams):

Param	Type	Description
model	string	Target model key on Console (required)
client	ConsoleClient | ConsoleClientOptions	An SDK client or options to build one (required)
temperature, topP, maxTokens, presencePenalty, frequencyPenalty, toolChoice	—	Defaults applied when a call omits them

Streaming (model.stream(...)) and tool calls are supported through the standard LangChain interfaces.

Tracing callback handler

CognipeerTracingCallbackHandler converts LangChain run events into a Cognipeer tracing session. Attach it via the callbacks array.

import { CognipeerTracingCallbackHandler } from '@cognipeer/console-sdk';

const tracer = new CognipeerTracingCallbackHandler({
  client: { apiKey: process.env.COGNIPEER_API_KEY! },
  agent: { name: 'support-agent' },
  threadId: 'conversation_42',
  autoFlush: true,
});

await chain.invoke(input, { callbacks: [tracer] });

Options (CognipeerTracingCallbackOptions):

Option	Description
client	SDK client or options (required)
sessionId	Session id (autogenerated if omitted)
threadId	Groups multiple sessions into one thread
agent	Agent descriptor stored with the session
summary, config	Initial summary / config blob stored on the session
autoFlush	Flush events on chain/tool completion
debug, logger	Verbose logging hooks

Helpers

• createCognipeerAgentTracing(options) — returns a binding that wires a tracing handler to an agent run with the same options as above.
• createCognipeerTracingMiddleware(...) — produces middleware for frameworks that take middleware rather than a callbacks array.

const tracing = createCognipeerAgentTracing({
  client: { apiKey: process.env.COGNIPEER_API_KEY! },
  agent: { name: 'support-agent' },
});

For graph-based agents see the LangGraph integration; for vendor-neutral spans see OpenTelemetry.