The message queue built for AI workflows.

DriftQ is the AI-native reliability layer for agent workflows. DriftQ-Core is the broker you can run today. See the roadmap for what’s next.

DriftQ-Core is the repo you run today.

Run locally View Docs

Alpha — not production-ready

Shipped:DriftQ-Core MVP

Shipped:Docker images, docs polish, etc.

Now:Replayable Workflow Runtime

What you can build with DriftQ

DriftQ is not “just a queue”. It’s the reliability layer for systems that need retries, backoff, idempotency, and observability — without fragile glue code.

• Agent pipelines that fan-out work and safely retry failures
• Long-running workflows with durable steps and strict DLQ handling
• Streaming consumers that can crash and recover without double-processing
• Backpressure-aware producers that get explicit 429 + Retry-After

See real-world use cases

Concrete scenarios: Next.js + FastAPI + LangChain pipelines, retries, DLQ, replay, and more.

Example

# Create a topic
curl -i -X POST "http://localhost:8080/v1/topics?name=t&partitions=1"

# Produce with a retry policy
curl -i -X POST "http://localhost:8080/v1/produce?topic=t&value=hello&retry_max_attempts=2"

# Stream consume (NDJSON)
curl --no-buffer "http://localhost:8080/v1/consume?topic=t&group=g&owner=o&lease_ms=5000"

# Ack a message
curl -i -X POST "http://localhost:8080/v1/ack?topic=t&group=g&owner=o&partition=0&offset=0"

# Metrics
curl -s "http://localhost:8080/metrics" | findstr consumer_lag

Core capabilities

The primitives you need for safe, retryable AI work.

Streaming consume

NDJSON streaming from /v1/consume with per-owner leases.

Retry + DLQ

Automatic redelivery, envelope retry policy, and strict DLQ routing.

Idempotency

Consume-scope idempotency keys to prevent duplicate side effects.

Observability

Prometheus metrics for inflight, lag, DLQ totals, and backpressure rejects.