Your AI features,
traced and priced.
Every LLM call your app makes, with its model, tokens, estimated cost, prompt, and response. Follow an agent loop step by step, right next to your database queries.
Per model, cache and reasoning
Follow an agent loop end to end
Evaluation scores, explained
For the MCP server you ship
One integration, every call traced
Add the Vercel AI SDK or Mastra integration to your config, and every model call becomes a span with its model, tokens, and timing. No wrapper around your calls.
- generateText, streamText, generateObject, streamObject, embed, embedMany, and rerank
- Follows the OpenTelemetry GenAI conventions, so any instrumentation that emits them works
- Leave prompts and responses out of your spans when they hold sensitive data
import { defineConfig } from '@monocle.sh/adonisjs-agent'
import { vercelAi } from '@monocle.sh/instrumentation-vercel-ai/integration'
import env from '#start/env'
export default defineConfig({
dsn: env.get('MONOCLE_DSN'),
integrations: [vercelAi({ recordInputs: false, recordOutputs: false })],
})Know what each conversation costs
The AI Agents dashboard breaks your calls down by model and tool: volume, latency, errors, tokens, and estimated cost. Group the calls of a multi-turn interaction into a conversation with one attribute.
- Input, output, cached, cache write, and reasoning tokens per call
- Costs estimated from published model pricing
- Tool call stats with success and error rates, and scorer results with their explanations
Follow an agent loop, step by step
AI calls are spans like any other. Open the trace of a request and see every model call and tool call of the agent, next to your database queries and HTTP calls.
- An AI tab on each LLM span with the model, tokens, prompt, response, and tool calls
- A slow answer? See whether the model, a tool, or your own code took the time
- Pipelines named after their function ID, and streaming calls marked as such
Monitor the MCP server you ship
If your app exposes an MCP server, Monocle shows how AI clients use it: which tools are called, by which clients and users, and how fast.
- Tool calls, P95 latency, error rate, unique users, and clients
- Per-tool detail with recent calls linked to their traces
- Supports the MCP TypeScript SDK v1 and v2
Plugs into your stack
Vercel AI SDK
generateText, streamText, and more
Mastra
Agents and workflows
MCP SDK
v1 and v2 servers
OpenTelemetry GenAI
Any instrumentation that follows it
AdonisJS
One line in config/monocle.ts
Traces & errors
AI spans in the same waterfall
See what your apps
are really doing.
Traces, logs, errors and metrics for every app you ship. Bring it all into focus.