Connect in two minutes.

Sign up, create a project for one AI app and describe what it should do. Jev checks every trace against that. You get an ingest token that starts with sut_. It is shown once, so keep it in your secret manager.

In the project, choose Connect → Langfuse and enter the Langfuse project’s public key, secret key and host (cloud.langfuse.com or your own). Keys are stored encrypted.

Suture backfills the last 7 days, then polls every minute. It backs off when Langfuse rate-limits, and never ingests the same trace twice.

Point your OTLP/HTTP exporter at Suture. The endpoint accepts protobuf and JSON, and reads the OpenTelemetry GenAI and OpenInference attribute names, so most LLM instrumentation libraries work as they are.

OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=https://<your-suture-host>/api/otlp/v1/traces
OTEL_EXPORTER_OTLP_TRACES_HEADERS=Authorization=Bearer%20<ingest-token>
OTEL_EXPORTER_OTLP_TRACES_PROTOCOL=http/protobuf
OTEL_SERVICE_NAME=content-api

from opentelemetry import trace
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import BatchSpanProcessor
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter

provider = TracerProvider()
provider.add_span_processor(BatchSpanProcessor(OTLPSpanExporter()))  # reads OTEL_* env vars
trace.set_tracer_provider(provider)

import { NodeSDK } from '@opentelemetry/sdk-node';
import { OTLPTraceExporter } from '@opentelemetry/exporter-trace-otlp-proto';

// Reads OTEL_* env vars. Start before the rest of your app loads.
new NodeSDK({ traceExporter: new OTLPTraceExporter() }).start();

A fix needs to know which file a prompt lives in. Tag the span or the code, let Suture match prompt text against your repo, or map it once by hand in the app.

# On the span that calls the model
span.set_attribute("suture.prompt_id", "orchestrator.system")

// Or next to the prompt in your code
// @suture prompt:orchestrator.system

Install the Suture GitHub App on the repository, pick an autonomy level (the default is pr), and Suture opens one small, proven PR per Issue. Nothing merges until you approve.

The MCP server lets Claude Code or Cursor list Issues, read fix plans and pull failing cases into your session.

claude mcp add --transport http suture https://<your-suture-host>/api/mcp \
  --header "Authorization: Bearer <personal-token>"