Skip to main content
Moda is a harness engineering platform: it turns production traces into verified improvements for your agent harness (prompts, tools, skills, evals, memory). A trace is the full record of one agent run — every message, tool call, thinking block, and step — whether or not a human is in the loop. Ingestion is observability-grade — you send traces through the Python or Node.js SDK, OpenTelemetry, or the HTTP Ingestion API. Moda analyzes every trace automatically — grouping it into use-case clusters and detecting user emotion, tool failure, laziness, and hallucination signals, which are rolled up into problems — and you explore the results in the dashboard or query them with the Data API and CLI.

​
The three surfaces

​
Ingestion

The Ingestion API accepts trace data from the Moda SDKs for Python and Node.js, which auto-instrument the OpenAI and Anthropic client libraries, and from any OpenTelemetry setup that exports OTLP over HTTP, or a direct HTTP call. Integrations also cover the Vercel AI SDK, the Claude Agent SDK, coding agents, and providers such as OpenRouter, Azure OpenAI, Bedrock, Vapi, and OpenClaw. Start with the ingestion overview or the quickstart.

​
Dashboard

The dashboard at moda.dev/dashboard is where you read the analysis: browse traces with a full timeline and world state, review signals and problems, and explore use cases, users, and tools. The dashboard overview describes each section.

​
Data API & CLI

Everything the dashboard shows is also queryable. The Data API is an HTTP API at https://moda.dev/api/v1/data, authenticated with your API key. The CLI (@moda-ai/cli, binary moda) brings search, analytics, signal and Problem investigation (emotions, hallucinations, Problem dossiers, step scores, world state), live tailing, and fix workflows to your terminal and CI. The hosted MCP server exposes the same commands as MCP tools, so Claude Code, Cursor, or any MCP client can query your production data directly.

​
How data flows

  1. Your application sends traces to the Ingestion API — through an SDK, an OpenTelemetry exporter, or a direct HTTP request.
  2. Moda processes new data within minutes; traces appear on the Traces page shortly after ingest.
  3. Emotion, tool-failure, laziness, and hallucination signals appear within minutes, and problems and use-case clusters build automatically as evidence accumulates.

​
Getting access

Create an account at moda.dev/early-access-2026q2. Your first sign-in creates a workspace and starts onboarding. Teammates join an existing workspace by invitation (Settings → Members & roles). To send data you need an API key (called an ingestion key in the dashboard): create one at Settings → Ingestion keys. The full key is shown once, at creation.

​
Next steps