synapseorch-ai/synapse-ai](https:

v1.7.0 — Scale Layer: Run Any Number of Agents or Orchestrations, Concurrently

This is the biggest release since Synapse launched. v1.7.0 ships the distributed scale layer — a complete production infrastructure that lets you go from a single-user local tool to a platform handling millions of orchestration runs per month, without changing a single line of application code that calls the API.

What Changed

Standalone mode executes workflows in the same process that serves the API. One crash, everything goes down. One long pipeline, everything else waits.

Scale mode separates the concerns entirely:

• API servers enqueue to Redis and return a run_id in under 1 ms
• Independent worker processes pull from Redis and execute workflows
• Workers publish real-time events back over Redis Streams (SSE)
• PostgreSQL holds full run state with per-step checkpoints
• S3 stores large artifacts so Postgres stays lean

Add more workers and throughput scales linearly. The API surface never changes.

New in This Release

Scale Infrastructure

• ARQ worker fleet — pull-based Python workers, each running 1–20 concurrent orchestrations. Start one or a hundred. They share nothing except Redis and Postgres.
• Redis job queue + SSE streams — jobs enqueue in under 1 ms, stream events survive client reconnects via Last-Event-ID resume
• Per-step Postgres checkpoints — a worker crash mid-run doesn't lose the job. The next worker picks it up from the last completed step.
• Dead-letter queue — failed jobs land in a DLQ with full error tracebacks, attempt counts, and one-click retry from the UI
• Stale worker reaper — workers that crash hard (SIGKILL, OOM) are detected and marked offline automatically after 90 seconds of silence
• Multi-tenant quotas — assign each team or customer isolated concurrent run limits; excess requests return HTTP 429 before touching the queue

S3 Artifact Storage

• Large file outputs, document attachments, and batch results stream directly to AWS S3, Cloudflare R2, or self-hosted MinIO
• New file_reader tool that reads from local filesystem or S3 — agents can reference S3 objects transparently
• S3 connection test built into the Settings UI

V2 API

• Stable versioned REST API at /api/v2/ for building products on top of Synapse
• POST .../run → enqueue and get run_id immediately (HTTP 202)
• GET .../stream → SSE event stream for real-time step-by-step updates
• GET .../status → poll-based status for mobile / webhook consumers
• POST .../cancel → distributed cancellation via Redis signal
• Webhook callbacks with HMAC signature verification
• Prometheus metrics at /api/v2/metrics
• OpenTelemetry tracing via OTLP_ENDPOINT

Settings → Scale Tab

• Full UI for configuring Redis URL, Postgres URL, and S3 credentials — one-click Test for each connection
• Live worker list with status, active job count, and last heartbeat
• Queue depth stats, DLQ inspector with expandable error views and retry button
• One-click Sync Now to push your orchestration and agent definitions to Postgres

Kubernetes & Observability

• Production-ready K8s manifests in infra/k8s/: API deployment, worker deployment, PgBouncer deployment
• KEDA ScaledObject — workers autoscale on Redis queue depth (1 replica per N queued jobs, 1–100 range)
• Prometheus scrape config in infra/prometheus.yml

Docker Images

Three separate images are now published — pull only what you need:

# Full app (API + frontend, standalone mode)
docker pull synapseorchai/synapse-ai:1.7.0

# Stateless API server only (scale mode)
docker pull synapseorchai/synapse-ai-api-server:1.7.0

# Worker process (scale mode, run as many as you need)
docker pull synapseorchai/synapse-ai-worker:1.7.0