vequil

Prevent runaway agents. Detect anomalies. Track cost.

Reliability and observability infrastructure for AI agent operators. Connect any runtime — every action is logged, anomalies are flagged in real time, and spend is tracked by agent and tool.

npm install vequil

Quickstart

import { VequilClient } from 'vequil'

const vq = new VequilClient({ workspaceKey: 'vk_ws_...' })

// Log any agent event
await vq.ingest({
  source: 'my-agent',
  eventType: 'tool_call',
  eventStatus: 'success',
  agentId: 'ops-agent-1',
  sessionId: 'session-abc',
  toolName: 'bash',
  costUsd: 0.012,
})

// Wrap any async call — logs success or failure automatically
const result = await vq.watch(
  { source: 'openclaw', agentId: 'ops-agent-1', toolName: 'bash' },
  () => runBashCommand('ls -la'),
)

Integrations

LangChain

import { VequilCallbackHandler } from 'vequil/integrations/langchain'

const handler = new VequilCallbackHandler(vq, { agentId: 'my-chain' })
await chain.invoke({ input: '...' }, { callbacks: [handler] })

Tool calls, LLM invocations, and chain errors are all captured automatically.

LangGraph

import { vequilNode } from 'vequil/integrations/langgraph'

const searchNode = vequilNode(
  vq,
  { agentId: 'my-graph', toolName: 'search' },
  async (state) => { /* your node logic */ return state },
)
graph.addNode('search', searchNode)

Vercel AI SDK

import { withVequil } from 'vequil/integrations/vercel-ai'
import { generateText } from 'ai'
import { openai } from '@ai-sdk/openai'

const trackedGenerate = withVequil(vq, { agentId: 'my-agent' }, generateText)
const { text } = await trackedGenerate({ model: openai('gpt-4o'), prompt: '...' })

OpenTelemetry (Vercel AI SDK, LangChain JS, any OTel-compatible runtime)

Drop-in span exporter. Works with Vercel AI SDK's experimental_telemetry and any other OTel-instrumented framework.

import { NodeTracerProvider } from '@opentelemetry/sdk-trace-node'
import { SimpleSpanProcessor } from '@opentelemetry/sdk-trace-base'
import { VequilSpanExporter } from 'vequil/integrations/otel'

const provider = new NodeTracerProvider()
provider.addSpanProcessor(new SimpleSpanProcessor(new VequilSpanExporter(vq)))
provider.register()

// Vercel AI SDK spans are now captured automatically
const { text } = await generateText({
  model: openai('gpt-4o'),
  prompt: 'Hello',
  experimental_telemetry: { isEnabled: true, functionId: 'my-agent' },
})

Environment variables

| Variable | Description | |---|---| | VEQUIL_WORKSPACE_KEY | Your workspace ingest key (alternative to passing in constructor) | | VEQUIL_URL | Custom instance URL (default: https://vequil.com) |

Get your workspace key at vequil.com/quickstart.

What gets detected

• Logic loops and retry storms
• Runaway spend (cost spikes by agent or tool)
• Failed tool calls and chain errors
• Missing auth keys and blocked actions
• Session-level failure chains (cascade detection)