DriftQ logo
DriftQ
Open-source • Developer-first •BETA• Durable broker + governed runtime

A durable broker and replayable AI runtime for governed agent orchestration.

DriftQ combines durable messaging with replayable workflows, policy and risk checks, human-in-the-loop approvals, secure tool execution, agent memory, and production-ready observability in one Go core.

Run LocallyView Docs
Replay + forensicsPolicy + HITLSecure tool gatewayOTel + metricsMulti-agent runtime
Quickstart
Run the server, verify the broker, then jump into the runtime docs.
driftqctl
1) Run
docker
docker run --rm \
-p 8080:8080 \
-v driftq-data:/data \
ghcr.io/driftq-org/driftq-core:1.3.0
2) Create topic
CLI

driftqctl topics create --name demo --partitions 1

3) Produce + consume
streaming
# produce
curl -X POST http://localhost:8080/v1/produce \
-H "content-type: application/json" \
-d '{"topic":"demo","value":"hello"}'

# consume (stream)
driftqctl topics peek --topic demo --group g1
The same server also exposes replay, policy, eval, agent-state, and runtime debug routes.
See the full guide ?

Embedded dashboard, same binary

DriftQ-Core ships a built-in dashboard under /ui/. No separate service, no extra deploy, and no detached admin app to keep in sync.

  • Served by the same driftqd process as the API
  • Included in the Docker image and local Docker flow
  • Covers overview, topics, runs, artifacts, runtime state, and debug surfaces

Run DriftQ, then open http://localhost:8080/ui/.

Built-in dashboard preview
Imported from the current DriftQ-Core repo UI assets
/ui/
DriftQ embedded dashboard UI showing overview metrics, topics, and live events

Runtime primitives that do not leak complexity

You should not have to rebuild durability, replay, governance, tool safety, and observability every time you ship an AI workflow. DriftQ keeps the hard parts explicit and the happy path fast.

  • Durable topics + WAL-backed storage
  • Replay, lineage, and what-if branch timelines
  • Policy, risk, HITL, and tenant governance
  • Tool gateway, receipts, and OpenTelemetry-native telemetry
Replay + forensics

Time-travel replay, run diffs, workflow diffs, root-cause views, and branching what-if simulations.

Guardrails + HITL

RBAC, policy checks, runtime risk scoring, approvals, edits, timeouts, and governed resume flows.

Secure tool execution

Approved tool registry, schema validation, secret redaction, tool-call audit logs, and staged side effects.

Observability

OTel traces, runtime metrics, broker telemetry, and traceable run, node, tool, and approval spans.

Failure is normal. Make recovery boring.

DriftQ is designed for workflows that touch flaky downstreams: LLMs, tools, third-party APIs, webhooks, and long-running jobs. The system should recover with evidence, not just dump the problem on your on-call.

multi-agent taskstool callsapprovalsworkflow releasesreplay branches

Want a deeper tour? Start with Use Cases.

Replayable runtime
Guardrails + governance
Secure tool calls
Observability
Workflow execution that stays inspectable

DriftQ keeps runs replayable and governed, so you can pause for humans, inspect lineage, branch a replay, or stage risky side effects without rebuilding runtime plumbing yourself.

Replay + lineage

Re-drive from the step that changed, compare alternate branches, and inspect what changed across runs.

Guardrails

Policy, risk scoring, HITL approvals, tenant boundaries, and secure tool execution are first-class.

Observe in real time
OTel + metrics
workflow_runs_totalstatus=completed tenant=acme
human_tasks_waitingqueue=review
tool_call_duration_mstool=send_email route=primary

Core telemetry is OpenTelemetry-native, so traces and metrics can flow into the rest of your stack.

Ship governed AI workflows without rebuilding your runtime stack.

If you already feel the pain of retries, tool safety, replay, runtime visibility, or human approvals, DriftQ is worth a serious look. If not, keep things simple until you need the trust layer.

Quickstart